COMPMID-3639: (3RDPARTY_UPDATE) Move CL kernels to src

Change-Id: I10d27db788e5086adae1841e3e2441cd9b76ef84
Signed-off-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4310
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

121 files changed, 1118 insertions, 670 deletions

diff --git a/arm_compute/runtime/CL/ICLOperator.h b/arm_compute/runtime/CL/ICLOperator.h
index 526b7e93e9..38bcaf32f2 100644
--- a/arm_compute/runtime/CL/ICLOperator.h
+++ b/arm_compute/runtime/CL/ICLOperator.h
@@ -24,7 +24,8 @@
 #ifndef ARM_COMPUTE_ICLOPERATOR_H
 #define ARM_COMPUTE_ICLOPERATOR_H
 
-#include "arm_compute/core/CL/ICLKernel.h"
+#include "arm_compute/core/Types.h"
+
 #include "arm_compute/runtime/IOperator.h"
 #include "arm_compute/runtime/IRuntimeContext.h"
 #include "arm_compute/runtime/Types.h"
@@ -33,6 +34,7 @@
 
 namespace arm_compute
 {
+class ICLKernel;
 namespace experimental
 {
 /** Basic interface for functions which have a single async CL kernel */
diff --git a/arm_compute/runtime/CL/ICLSimpleFunction.h b/arm_compute/runtime/CL/ICLSimpleFunction.h
index 4b1d5b1485..310bf770c4 100644
--- a/arm_compute/runtime/CL/ICLSimpleFunction.h
+++ b/arm_compute/runtime/CL/ICLSimpleFunction.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2019 Arm Limited.
+ * Copyright (c) 2016-2020 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_ICLSIMPLEFUNCTION_H
 #define ARM_COMPUTE_ICLSIMPLEFUNCTION_H
 
-#include "arm_compute/core/CL/ICLKernel.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <memory>
@@ -34,6 +32,8 @@ namespace arm_compute
 {
 // Forward declarations
 class CLRuntimeContext;
+class CLFillBorderKernel;
+class ICLKernel;
 
 /** Basic interface for functions which have a single OpenCL kernel */
 class ICLSimpleFunction : public IFunction
@@ -53,14 +53,16 @@ public:
     ICLSimpleFunction &operator=(const ICLSimpleFunction &) = delete;
     /** Default move assignment operator */
     ICLSimpleFunction &operator=(ICLSimpleFunction &&) = default;
+    /** Default destructor */
+    ~ICLSimpleFunction();
 
     // Inherited methods overridden:
     void run() override final;
 
 protected:
-    std::unique_ptr<ICLKernel> _kernel;         /**< Kernel to run */
-    CLFillBorderKernel         _border_handler; /**< Kernel to handle  borders */
-    CLRuntimeContext          *_ctx;            /**< Context to use */
+    std::unique_ptr<ICLKernel>          _kernel;         /**< Kernel to run */
+    std::unique_ptr<CLFillBorderKernel> _border_handler; /**< Kernel to handle  borders */
+    CLRuntimeContext                   *_ctx;            /**< Context to use */
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_ICLSIMPLEFUNCTION_H */
diff --git a/arm_compute/runtime/CL/functions/CLAbsoluteDifference.h b/arm_compute/runtime/CL/functions/CLAbsoluteDifference.h
index b0f1948beb..f2831e2a99 100644
--- a/arm_compute/runtime/CL/functions/CLAbsoluteDifference.h
+++ b/arm_compute/runtime/CL/functions/CLAbsoluteDifference.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLAbsoluteDifferenceKernel
diff --git a/arm_compute/runtime/CL/functions/CLAccumulate.h b/arm_compute/runtime/CL/functions/CLAccumulate.h
index 9dbf13b873..20d3476d2e 100644
--- a/arm_compute/runtime/CL/functions/CLAccumulate.h
+++ b/arm_compute/runtime/CL/functions/CLAccumulate.h
@@ -30,6 +30,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLAccumulateKernel */
diff --git a/arm_compute/runtime/CL/functions/CLActivationLayer.h b/arm_compute/runtime/CL/functions/CLActivationLayer.h
index 632487c78d..dc2cb62b71 100644
--- a/arm_compute/runtime/CL/functions/CLActivationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLActivationLayer.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLActivationLayerKernel
  *
diff --git a/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h b/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
index dc0c37e860..c254284cd7 100644
--- a/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
+++ b/arm_compute/runtime/CL/functions/CLArgMinMaxLayer.h
@@ -24,7 +24,6 @@
 #ifndef ARM_COMPUTE_CLARGMINMAXLAYER_H
 #define ARM_COMPUTE_CLARGMINMAXLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLArgMinMaxLayerKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
@@ -36,6 +35,7 @@ namespace arm_compute
 {
 class ITensorInfo;
 class ICLTensor;
+class CLArgMinMaxLayerKernel;
 
 /** Function to calculate the index of the minimum or maximum values in a
  *  tensor based on an axis.
@@ -53,6 +53,16 @@ public:
      * @param[in] memory_manager (Optional) Memory manager.
      */
     CLArgMinMaxLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLArgMinMaxLayer(const CLArgMinMaxLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLArgMinMaxLayer &operator=(const CLArgMinMaxLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLArgMinMaxLayer(CLArgMinMaxLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLArgMinMaxLayer &operator=(CLArgMinMaxLayer &&) = delete;
+    /** Default destructor */
+    ~CLArgMinMaxLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input  Input source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/S32/F16/F32.
@@ -85,13 +95,13 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                         _memory_group;
-    std::vector<CLTensor>               _results_vector;
-    CLTensor                            _not_reshaped_output;
-    std::vector<CLArgMinMaxLayerKernel> _reduction_kernels_vector;
-    CLReshapeLayer                      _reshape;
-    unsigned int                        _num_of_stages;
-    unsigned int                        _reduction_axis;
+    MemoryGroup                                          _memory_group;
+    std::vector<CLTensor>                                _results_vector;
+    CLTensor                                             _not_reshaped_output;
+    std::vector<std::unique_ptr<CLArgMinMaxLayerKernel>> _reduction_kernels_vector;
+    CLReshapeLayer                                       _reshape;
+    unsigned int                                         _num_of_stages;
+    unsigned int                                         _reduction_axis;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLARGMINMAXLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLBatchNormalizationLayer.h b/arm_compute/runtime/CL/functions/CLBatchNormalizationLayer.h
index c22991da7c..c8acf9fc6b 100644
--- a/arm_compute/runtime/CL/functions/CLBatchNormalizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLBatchNormalizationLayer.h
@@ -26,12 +26,16 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLBatchNormalizationLayerKernel.h"
 #include "arm_compute/core/Types.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
+class CLBatchNormalizationLayerKernel;
 
 /** Basic function to run @ref CLNormalizationLayerKernel and simulate a batch normalization layer.
  *
@@ -44,6 +48,16 @@ class CLBatchNormalizationLayer : public IFunction
 public:
     /** Default constructor */
     CLBatchNormalizationLayer();
+    /** Prevent instances of this class from being copied */
+    CLBatchNormalizationLayer(const CLBatchNormalizationLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLBatchNormalizationLayer &operator=(const CLBatchNormalizationLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLBatchNormalizationLayer(CLBatchNormalizationLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLBatchNormalizationLayer &operator=(CLBatchNormalizationLayer &&) = delete;
+    /** Default destructor */
+    ~CLBatchNormalizationLayer();
     /** Set the input and output tensors.
      *
      * @note If the output tensor is a nullptr or is equal to the input, the batch normalization function will be performed in-place
@@ -104,7 +118,7 @@ public:
     void run() override;
 
 private:
-    CLBatchNormalizationLayerKernel _norm_kernel; /**< BatchNormalization layer kernel to run */
+    std::unique_ptr<CLBatchNormalizationLayerKernel> _norm_kernel; /**< BatchNormalization layer kernel to run */
 };
-}
+} // namespace arm_compute
 #endif /* ARM_COMPUTE_CLBATCHNORMALIZATIONLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLBatchToSpaceLayer.h b/arm_compute/runtime/CL/functions/CLBatchToSpaceLayer.h
index ba57921cc2..bdb58531d0 100644
--- a/arm_compute/runtime/CL/functions/CLBatchToSpaceLayer.h
+++ b/arm_compute/runtime/CL/functions/CLBatchToSpaceLayer.h
@@ -26,11 +26,15 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLBatchToSpaceLayerKernel.h"
 #include "arm_compute/core/Types.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class ITensorInfo;
+class CLBatchToSpaceLayerKernel;
 class ICLTensor;
 
 /** Basic function to run @ref CLBatchToSpaceLayerKernel. */
@@ -39,6 +43,16 @@ class CLBatchToSpaceLayer : public IFunction
 public:
     /** Default constructor */
     CLBatchToSpaceLayer();
+    /** Prevent instances of this class from being copied */
+    CLBatchToSpaceLayer(const CLBatchToSpaceLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLBatchToSpaceLayer &operator=(const CLBatchToSpaceLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLBatchToSpaceLayer(CLBatchToSpaceLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLBatchToSpaceLayer &operator=(CLBatchToSpaceLayer &&) = delete;
+    /** Default destructor */
+    ~CLBatchToSpaceLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input       Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -95,7 +109,7 @@ public:
     void run() override;
 
 private:
-    CLBatchToSpaceLayerKernel _batch_to_space_kernel; /**< CLBatchToSpaceLayerKernel to run */
+    std::unique_ptr<CLBatchToSpaceLayerKernel> _batch_to_space_kernel; /**< CLBatchToSpaceLayerKernel to run */
 };
-}
+} // namespace arm_compute
 #endif /* ARM_COMPUTE_CLBATCHTOSPACELAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLBitwiseAnd.h b/arm_compute/runtime/CL/functions/CLBitwiseAnd.h
index 3c28938807..bf5993f4b0 100644
--- a/arm_compute/runtime/CL/functions/CLBitwiseAnd.h
+++ b/arm_compute/runtime/CL/functions/CLBitwiseAnd.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLBitwiseAndKernel.
diff --git a/arm_compute/runtime/CL/functions/CLBitwiseNot.h b/arm_compute/runtime/CL/functions/CLBitwiseNot.h
index 4c21d5647f..1d8531a176 100644
--- a/arm_compute/runtime/CL/functions/CLBitwiseNot.h
+++ b/arm_compute/runtime/CL/functions/CLBitwiseNot.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLBitwiseNotKernel.
diff --git a/arm_compute/runtime/CL/functions/CLBitwiseOr.h b/arm_compute/runtime/CL/functions/CLBitwiseOr.h
index 8a481737e3..7876cbf196 100644
--- a/arm_compute/runtime/CL/functions/CLBitwiseOr.h
+++ b/arm_compute/runtime/CL/functions/CLBitwiseOr.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLBitwiseOrKernel.
diff --git a/arm_compute/runtime/CL/functions/CLBitwiseXor.h b/arm_compute/runtime/CL/functions/CLBitwiseXor.h
index 6928e59d38..4f054062cd 100644
--- a/arm_compute/runtime/CL/functions/CLBitwiseXor.h
+++ b/arm_compute/runtime/CL/functions/CLBitwiseXor.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLBitwiseXorKernel.
diff --git a/arm_compute/runtime/CL/functions/CLBoundingBoxTransform.h b/arm_compute/runtime/CL/functions/CLBoundingBoxTransform.h
index 5e4e89071b..d6409106da 100644
--- a/arm_compute/runtime/CL/functions/CLBoundingBoxTransform.h
+++ b/arm_compute/runtime/CL/functions/CLBoundingBoxTransform.h
@@ -24,12 +24,16 @@
 #ifndef ARM_COMPUTE_CLBOUNDINGBOXTRANSOFORM_H
 #define ARM_COMPUTE_CLBOUNDINGBOXTRANSOFORM_H
 
-#include "arm_compute/core/CL/kernels/CLBoundingBoxTransformKernel.h"
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLBoundingBoxTransformKernel;
+class BoundingBoxTransformInfo;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLBoundingBoxTransformKernel.
  *
diff --git a/arm_compute/runtime/CL/functions/CLBox3x3.h b/arm_compute/runtime/CL/functions/CLBox3x3.h
index 2d2aa4705c..cff780614c 100644
--- a/arm_compute/runtime/CL/functions/CLBox3x3.h
+++ b/arm_compute/runtime/CL/functions/CLBox3x3.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute box filter 3x3. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLCannyEdge.h b/arm_compute/runtime/CL/functions/CLCannyEdge.h
index f9d9f8f66a..9e41c31728 100644
--- a/arm_compute/runtime/CL/functions/CLCannyEdge.h
+++ b/arm_compute/runtime/CL/functions/CLCannyEdge.h
@@ -26,8 +26,6 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLCannyEdgeKernel.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
@@ -36,6 +34,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLGradientKernel;
+class CLEdgeNonMaxSuppressionKernel;
+class CLEdgeTraceKernel;
 class ICLTensor;
 
 /** Basic function to execute canny edge on OpenCL. This function calls the following OpenCL kernels and functions:
@@ -56,6 +59,8 @@ public:
     CLCannyEdge(const CLCannyEdge &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLCannyEdge &operator=(const CLCannyEdge &) = delete;
+    /** Default destructor */
+    ~CLCannyEdge();
     /** Initialise the function's source, destination, thresholds, gradient size, normalization type and border mode.
      *
      * @param[in,out] input                 Source tensor. Data types supported: U8. (Written to only for border_mode != UNDEFINED)
@@ -88,20 +93,20 @@ public:
     virtual void run() override;
 
 private:
-    MemoryGroup                   _memory_group;                                    /**< Function's memory group */
-    std::unique_ptr<IFunction>    _sobel;                                           /**< Pointer to Sobel kernel. */
-    CLGradientKernel              _gradient;                                        /**< Gradient kernel. */
-    CLFillBorderKernel            _border_mag_gradient;                             /**< Fill border on magnitude tensor kernel */
-    CLEdgeNonMaxSuppressionKernel _non_max_suppr;                                   /**< Non-Maxima suppression kernel. */
-    CLEdgeTraceKernel             _edge_trace;                                      /**< Edge tracing kernel. */
-    CLImage                       _gx;                                              /**< Source tensor - Gx component. */
-    CLImage                       _gy;                                              /**< Source tensor - Gy component. */
-    CLImage                       _mag;                                             /**< Source tensor - Magnitude. */
-    CLImage                       _phase;                                           /**< Source tensor - Phase. */
-    CLImage                       _nonmax;                                          /**< Source tensor - Non-Maxima suppressed. */
-    CLImage                       _visited, _recorded, _l1_list_counter, _l1_stack; /**< Temporary tensors */
-    ICLTensor                    *_output;                                          /**< Output tensor provided by the user. */
+    MemoryGroup                                    _memory_group;                                    /**< Function's memory group */
+    std::unique_ptr<IFunction>                     _sobel;                                           /**< Pointer to Sobel kernel. */
+    std::unique_ptr<CLGradientKernel>              _gradient;                                        /**< Gradient kernel. */
+    std::unique_ptr<CLFillBorderKernel>            _border_mag_gradient;                             /**< Fill border on magnitude tensor kernel */
+    std::unique_ptr<CLEdgeNonMaxSuppressionKernel> _non_max_suppr;                                   /**< Non-Maxima suppression kernel. */
+    std::unique_ptr<CLEdgeTraceKernel>             _edge_trace;                                      /**< Edge tracing kernel. */
+    CLImage                                        _gx;                                              /**< Source tensor - Gx component. */
+    CLImage                                        _gy;                                              /**< Source tensor - Gy component. */
+    CLImage                                        _mag;                                             /**< Source tensor - Magnitude. */
+    CLImage                                        _phase;                                           /**< Source tensor - Phase. */
+    CLImage                                        _nonmax;                                          /**< Source tensor - Non-Maxima suppressed. */
+    CLImage                                        _visited, _recorded, _l1_list_counter, _l1_stack; /**< Temporary tensors */
+    ICLTensor                                     *_output;                                          /**< Output tensor provided by the user. */
 };
-}
+} // namespace arm_compute
 
 #endif /* ARM_COMPUTE_CLCANNYEDGE_H */
diff --git a/arm_compute/runtime/CL/functions/CLCast.h b/arm_compute/runtime/CL/functions/CLCast.h
index 592368d135..bd333d4e72 100644
--- a/arm_compute/runtime/CL/functions/CLCast.h
+++ b/arm_compute/runtime/CL/functions/CLCast.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLDepthConvertLayerKernel. */
 class CLCast : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLChannelCombine.h b/arm_compute/runtime/CL/functions/CLChannelCombine.h
index 4e3d10cc10..5927662fc2 100644
--- a/arm_compute/runtime/CL/functions/CLChannelCombine.h
+++ b/arm_compute/runtime/CL/functions/CLChannelCombine.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLMultiImage;
 class ICLTensor;
 using ICLImage = ICLTensor;
diff --git a/arm_compute/runtime/CL/functions/CLChannelExtract.h b/arm_compute/runtime/CL/functions/CLChannelExtract.h
index cf042b4519..9ce9bcdd8a 100644
--- a/arm_compute/runtime/CL/functions/CLChannelExtract.h
+++ b/arm_compute/runtime/CL/functions/CLChannelExtract.h
@@ -29,6 +29,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLMultiImage;
 class ICLTensor;
 using ICLImage = ICLTensor;
diff --git a/arm_compute/runtime/CL/functions/CLChannelShuffleLayer.h b/arm_compute/runtime/CL/functions/CLChannelShuffleLayer.h
index e0bb3d01c9..54cf59f59a 100644
--- a/arm_compute/runtime/CL/functions/CLChannelShuffleLayer.h
+++ b/arm_compute/runtime/CL/functions/CLChannelShuffleLayer.h
@@ -24,11 +24,14 @@
 #ifndef ARM_COMPUTE_CLCHANNELSHUFFLELAYER_H
 #define ARM_COMPUTE_CLCHANNELSHUFFLELAYER_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLChannelShuffleLayerKernel
  *
diff --git a/arm_compute/runtime/CL/functions/CLColorConvert.h b/arm_compute/runtime/CL/functions/CLColorConvert.h
index e4017c2686..47bcabfb63 100644
--- a/arm_compute/runtime/CL/functions/CLColorConvert.h
+++ b/arm_compute/runtime/CL/functions/CLColorConvert.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLMultiImage;
 class ICLTensor;
 using ICLImage = ICLTensor;
diff --git a/arm_compute/runtime/CL/functions/CLComparison.h b/arm_compute/runtime/CL/functions/CLComparison.h
index c6d61e45f2..8cc3e96ec5 100644
--- a/arm_compute/runtime/CL/functions/CLComparison.h
+++ b/arm_compute/runtime/CL/functions/CLComparison.h
@@ -30,7 +30,9 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLComparisonKernel */
 class CLComparison : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLComputeAllAnchors.h b/arm_compute/runtime/CL/functions/CLComputeAllAnchors.h
index a2f1a4eb66..d6a2ab423d 100644
--- a/arm_compute/runtime/CL/functions/CLComputeAllAnchors.h
+++ b/arm_compute/runtime/CL/functions/CLComputeAllAnchors.h
@@ -24,12 +24,15 @@
 #ifndef ARM_COMPUTE_CLCOMPUTEALLANCHORS_H
 #define ARM_COMPUTE_CLCOMPUTEALLANCHORS_H
 
-#include "arm_compute/core/CL/kernels/CLGenerateProposalsLayerKernel.h"
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
+class ComputeAnchorsInfo;
 
 /** Basic function to run @ref CLComputeAllAnchorsKernel.
  *
diff --git a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
index f535c8ea97..5e7003a112 100644
--- a/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
+++ b/arm_compute/runtime/CL/functions/CLConcatenateLayer.h
@@ -27,7 +27,6 @@
 #include "arm_compute/runtime/CL/ICLOperator.h"
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/ICLKernel.h"
 #include "arm_compute/core/Types.h"
 
 #include <memory>
@@ -36,7 +35,9 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
 class ICLTensor;
+class ICLKernel;
 class ITensorInfo;
 class Status;
 
diff --git a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
index 9298be2e53..75a3d3213e 100644
--- a/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
+++ b/arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h
@@ -24,14 +24,17 @@
 #ifndef ARM_COMPUTE_CLCONVERTFULLYCONNECTEDWEIGHTS_H
 #define ARM_COMPUTE_CLCONVERTFULLYCONNECTEDWEIGHTS_H
 
-#include "arm_compute/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 #include "arm_compute/runtime/ITransformWeights.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLConvertFullyConnectedWeightsKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLConvertFullyConnectedWeightsKernel. */
 class CLConvertFullyConnectedWeights : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLConvolution.h b/arm_compute/runtime/CL/functions/CLConvolution.h
index c06ad0d969..4a1631a702 100644
--- a/arm_compute/runtime/CL/functions/CLConvolution.h
+++ b/arm_compute/runtime/CL/functions/CLConvolution.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLCONVOLUTION_H
 #define ARM_COMPUTE_CLCONVOLUTION_H
 
-#include "arm_compute/core/CL/kernels/CLConvolutionKernel.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
@@ -38,6 +36,13 @@
 
 namespace arm_compute
 {
+template <unsigned int matrix_size>
+class CLConvolutionKernel;
+template <unsigned int matrix_size>
+class CLSeparableConvolutionHorKernel;
+template <unsigned int matrix_size>
+class CLSeparableConvolutionVertKernel;
+class CLFillBorderKernel;
 class ICLTensor;
 
 /** Basic function to execute convolution of size 3x3. This function calls the following OpenCL kernels:
@@ -85,6 +90,16 @@ class CLConvolutionSquare : public IFunction
 public:
     /** Default constructor */
     CLConvolutionSquare(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvolutionSquare(const CLConvolutionSquare &) = delete;
+    /** Default move constructor */
+    CLConvolutionSquare(CLConvolutionSquare &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvolutionSquare &operator=(const CLConvolutionSquare &) = delete;
+    /** Default move assignment operator */
+    CLConvolutionSquare &operator=(CLConvolutionSquare &&) = default;
+    /** Default destructor */
+    ~CLConvolutionSquare();
     /** Initialize the function's source, destination, conv and border_mode.
      *
      * @param[in,out] input                 Source tensor. Data types supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -111,13 +126,13 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                                   _memory_group;   /**< Function's memory group */
-    CLTensor                                      _tmp;            /**< temporary buffer for output of horizontal pass */
-    bool                                          _is_separable;   /**< true if the convolution can be separated */
-    CLSeparableConvolutionHorKernel<matrix_size>  _kernel_hor;     /**< kernel for horizontal pass of separated convolution */
-    CLSeparableConvolutionVertKernel<matrix_size> _kernel_vert;    /**< kernel for vertical pass of separated convolution */
-    CLConvolutionKernel<matrix_size>              _kernel;         /**< kernel for non-separated convolution **/
-    CLFillBorderKernel                            _border_handler; /**< kernel for border handling */
+    MemoryGroup                                                    _memory_group;   /**< Function's memory group */
+    CLTensor                                                       _tmp;            /**< temporary buffer for output of horizontal pass */
+    bool                                                           _is_separable;   /**< true if the convolution can be separated */
+    std::unique_ptr<CLSeparableConvolutionHorKernel<matrix_size>>  _kernel_hor;     /**< kernel for horizontal pass of separated convolution */
+    std::unique_ptr<CLSeparableConvolutionVertKernel<matrix_size>> _kernel_vert;    /**< kernel for vertical pass of separated convolution */
+    std::unique_ptr<CLConvolutionKernel<matrix_size>>              _kernel;         /**< kernel for non-separated convolution **/
+    std::unique_ptr<CLFillBorderKernel>                            _border_handler; /**< kernel for border handling */
 };
 
 /** Basic function to run 5x5 convolution. */
diff --git a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
index ac36523682..d1de721193 100644
--- a/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLConvolutionLayer.h
@@ -73,6 +73,16 @@ class CLConvolutionLayer : public IFunction
 public:
     /** Default constructor */
     CLConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Default Destructor */
+    ~CLConvolutionLayer();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvolutionLayer(const CLConvolutionLayer &) = delete;
+    /** Default move constructor */
+    CLConvolutionLayer(CLConvolutionLayer &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLConvolutionLayer &operator=(const CLConvolutionLayer &) = delete;
+    /** Default move assignment operator */
+    CLConvolutionLayer &operator=(CLConvolutionLayer &&) = default;
     /** Set the input and output tensors.
      *
      * @param[in]  input            Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
diff --git a/arm_compute/runtime/CL/functions/CLCopy.h b/arm_compute/runtime/CL/functions/CLCopy.h
index c20d75eea8..f1a091df84 100644
--- a/arm_compute/runtime/CL/functions/CLCopy.h
+++ b/arm_compute/runtime/CL/functions/CLCopy.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 class CLCopy : public ICLSimpleFunction
 {
diff --git a/arm_compute/runtime/CL/functions/CLCropResize.h b/arm_compute/runtime/CL/functions/CLCropResize.h
index e940928b90..e781cfe61f 100644
--- a/arm_compute/runtime/CL/functions/CLCropResize.h
+++ b/arm_compute/runtime/CL/functions/CLCropResize.h
@@ -25,9 +25,7 @@
 #define ARM_COMPUTE_CL_CROP_RESIZE_H
 
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/CL/kernels/CLCopyKernel.h"
-#include "arm_compute/core/CL/kernels/CLCropKernel.h"
-#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
+
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLScale.h"
 
@@ -37,7 +35,11 @@
 namespace arm_compute
 {
 // Forward Declarations
+class CLCompileContext;
+class CLCopyKernel;
+class CLCropKernel;
 class ITensor;
+class ITensorInfo;
 
 /** Function to perform cropping and resizing */
 class CLCropResize : public IFunction
@@ -54,7 +56,7 @@ public:
     /** Allow instances of this class to be moved */
     CLCropResize &operator=(CLCropResize &&) = default;
     /** Default destructor */
-    virtual ~CLCropResize() = default;
+    ~CLCropResize();
 
     /** Configure kernel
      *
diff --git a/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
index 19a44f7b93..3ebc858d32 100644
--- a/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
+++ b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
@@ -24,17 +24,20 @@
 #ifndef ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H
 #define ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H
 
-#include "arm_compute/runtime/IFunction.h"
-
-#include "arm_compute/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
-#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include <memory>
 
 namespace arm_compute
 {
 // Forward declarations
+class CLDeconvolutionLayerUpsampleKernel;
+class CLCompileContext;
+class CLMemsetKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute deconvolution upsample on OpenCL. This function calls the following OpenCL kernels and functions:
  *
@@ -55,7 +58,7 @@ public:
     /** Allow instances of this class to be moved */
     CLDeconvolutionLayerUpsample &operator=(CLDeconvolutionLayerUpsample &&) = default;
     /** Default destructor */
-    virtual ~CLDeconvolutionLayerUpsample() = default;
+    ~CLDeconvolutionLayerUpsample();
 
     /** Initialize the function's source, destination, interpolation type and border_mode.
      *
@@ -86,9 +89,9 @@ public:
     void run() override;
 
 private:
-    CLDeconvolutionLayerUpsampleKernel _upsample;
-    CLMemsetKernel                     _memset;
-    ICLTensor                         *_output;
+    std::unique_ptr<CLDeconvolutionLayerUpsampleKernel> _upsample;
+    std::unique_ptr<CLMemsetKernel>                     _memset;
+    ICLTensor                                          *_output;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H */
diff --git a/arm_compute/runtime/CL/functions/CLDepthConvertLayer.h b/arm_compute/runtime/CL/functions/CLDepthConvertLayer.h
index d125584c97..b0f297aec5 100644
--- a/arm_compute/runtime/CL/functions/CLDepthConvertLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDepthConvertLayer.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLDepthConvertLayerKernel. */
 class CLDepthConvertLayer : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLDepthToSpaceLayer.h b/arm_compute/runtime/CL/functions/CLDepthToSpaceLayer.h
index 5e197cb9b8..a0aa288dbf 100644
--- a/arm_compute/runtime/CL/functions/CLDepthToSpaceLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDepthToSpaceLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLDepthToSpaceLayerKernel. */
 class CLDepthToSpaceLayer : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h
index 558c4540fa..8e594bc09f 100644
--- a/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDepthwiseConvolutionLayer.h
@@ -24,12 +24,6 @@
 #ifndef ARM_COMPUTE_CLDEPTHWISECONVOLUTION_H
 #define ARM_COMPUTE_CLDEPTHWISECONVOLUTION_H
 
-#include "arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NCHWKernel.h"
-#include "arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayer3x3NHWCKernel.h"
-#include "arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayerNativeKernel.h"
-#include "arm_compute/core/CL/kernels/CLDepthwiseConvolutionLayerReshapeWeightsKernel.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/ICLDepthwiseConvolutionLayer3x3Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLPermute.h"
@@ -38,6 +32,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLDepthwiseConvolutionLayerNativeKernel;
+class CLDepthwiseConvolutionLayerReshapeWeightsKernel;
+class ICLDepthwiseConvolutionLayer3x3Kernel;
 class ICLTensor;
 
 /** Function to execute a depthwise convolution
@@ -55,6 +54,8 @@ public:
     CLDepthwiseConvolutionLayer &operator=(const CLDepthwiseConvolutionLayer &) = delete;
     /** Default move assignment operator */
     CLDepthwiseConvolutionLayer &operator=(CLDepthwiseConvolutionLayer &&) = default;
+    /** Default destructor */
+    ~CLDepthwiseConvolutionLayer();
     /** Initialize the function's source, destination, weights and convolution information.
      *
      * @param[in, out] input            Source tensor. Data type supported: QASYMM8/QASYMM8_SIGNED/FP16/FP32. Data layout supported: NHWC, NCHW
@@ -211,25 +212,25 @@ private:
         };
 
     private:
-        MemoryGroup                                            _memory_group;
-        std::unique_ptr<ICLDepthwiseConvolutionLayer3x3Kernel> _kernel;
-        CLFillBorderKernel                                     _border_handler;
-        CLPermute                                              _permute_input_to_nchw;
-        CLPermute                                              _permute_weights_to_nchw;
-        CLPermute                                              _permute_output_to_nhwc;
-        CLDepthwiseConvolutionLayerReshapeWeightsKernel        _reshape_weights;
-        CLTensor                                               _permuted_input;
-        CLTensor                                               _permuted_weights;
-        CLTensor                                               _permuted_output;
-        CLTensor                                               _output_multipliers;
-        CLTensor                                               _output_shifts;
-        const ITensor                                         *_original_weights;
-        const ITensor                                         *_input;
-        const ITensor                                         *_output;
-        bool                                                   _needs_permute;
-        bool                                                   _needs_weights_reshape;
-        bool                                                   _is_prepared;
-        bool                                                   _is_quantized;
+        MemoryGroup                                                      _memory_group;
+        std::unique_ptr<ICLDepthwiseConvolutionLayer3x3Kernel>           _kernel;
+        std::unique_ptr<CLFillBorderKernel>                              _border_handler;
+        CLPermute                                                        _permute_input_to_nchw;
+        CLPermute                                                        _permute_weights_to_nchw;
+        CLPermute                                                        _permute_output_to_nhwc;
+        std::unique_ptr<CLDepthwiseConvolutionLayerReshapeWeightsKernel> _reshape_weights;
+        CLTensor                                                         _permuted_input;
+        CLTensor                                                         _permuted_weights;
+        CLTensor                                                         _permuted_output;
+        CLTensor                                                         _output_multipliers;
+        CLTensor                                                         _output_shifts;
+        const ITensor                                                   *_original_weights;
+        const ITensor                                                   *_input;
+        const ITensor                                                   *_output;
+        bool                                                             _needs_permute;
+        bool                                                             _needs_weights_reshape;
+        bool                                                             _is_prepared;
+        bool                                                             _is_quantized;
     };
 
     /** Basic function to execute a generic depthwise convolution. This function calls the following OpenCL kernels:
@@ -313,10 +314,10 @@ private:
     private:
         MemoryGroup _memory_group;
 
-        CLDepthwiseConvolutionLayerNativeKernel _dwc_native_kernel;
-        CLPermute                               _permute_input_to_nhwc;
-        CLPermute                               _permute_weights_to_nhwc;
-        CLPermute                               _permute_output_to_nchw;
+        std::unique_ptr<CLDepthwiseConvolutionLayerNativeKernel> _dwc_native_kernel;
+        CLPermute                                                _permute_input_to_nhwc;
+        CLPermute                                                _permute_weights_to_nhwc;
+        CLPermute                                                _permute_output_to_nchw;
 
         CLTensor       _permuted_input;
         CLTensor       _permuted_weights;
diff --git a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
index 88ed915421..b2cf3356f4 100644
--- a/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDequantizationLayer.h
@@ -31,7 +31,9 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLDequantizationLayerKernel that dequantizes an input tensor */
 class CLDequantizationLayer : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLDerivative.h b/arm_compute/runtime/CL/functions/CLDerivative.h
index 1aba6a9f6c..4a91d5d50b 100644
--- a/arm_compute/runtime/CL/functions/CLDerivative.h
+++ b/arm_compute/runtime/CL/functions/CLDerivative.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute first order derivative operator. This function calls the following CL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLDilate.h b/arm_compute/runtime/CL/functions/CLDilate.h
index adb9cf4e6c..bf72cd3b26 100644
--- a/arm_compute/runtime/CL/functions/CLDilate.h
+++ b/arm_compute/runtime/CL/functions/CLDilate.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute dilate. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
index 8107fa24f3..0afc9d3f38 100644
--- a/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDirectConvolutionLayer.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYER_H
 #define ARM_COMPUTE_CLDIRECTCONVOLUTIONLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLDirectConvolutionLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -34,7 +32,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLDirectConvolutionLayerKernel;
+class CLFillBorderKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute direct convolution function:
  */
@@ -43,6 +45,12 @@ class CLDirectConvolutionLayer : public IFunction
 public:
     /** Default constructor */
     CLDirectConvolutionLayer();
+    /** Prevent instances of this class from being copied */
+    CLDirectConvolutionLayer(const CLDirectConvolutionLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLDirectConvolutionLayer &operator=(const CLDirectConvolutionLayer &) = delete;
+    /** Default destructor */
+    ~CLDirectConvolutionLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input     Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
@@ -95,9 +103,9 @@ public:
     void run() override;
 
 private:
-    CLDirectConvolutionLayerKernel _direct_conv_kernel;
-    CLFillBorderKernel             _input_border_handler;
-    CLActivationLayer              _activationlayer_function;
+    std::unique_ptr<CLDirectConvolutionLayerKernel> _direct_conv_kernel;
+    std::unique_ptr<CLFillBorderKernel>             _input_border_handler;
+    CLActivationLayer                               _activationlayer_function;
 
     bool _is_activationlayer_enabled;
 };
diff --git a/arm_compute/runtime/CL/functions/CLElementWiseUnaryLayer.h b/arm_compute/runtime/CL/functions/CLElementWiseUnaryLayer.h
index 5208bfe404..72b5b7dee8 100644
--- a/arm_compute/runtime/CL/functions/CLElementWiseUnaryLayer.h
+++ b/arm_compute/runtime/CL/functions/CLElementWiseUnaryLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to perform inverse square root on an input tensor. */
 class CLRsqrtLayer : public IFunction
diff --git a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
index 31d4f2e745..55c5fb3455 100644
--- a/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
+++ b/arm_compute/runtime/CL/functions/CLElementwiseOperations.h
@@ -30,6 +30,8 @@
 namespace arm_compute
 {
 class ICLTensor;
+class CLCompileContext;
+class ITensorInfo;
 
 namespace experimental
 {
diff --git a/arm_compute/runtime/CL/functions/CLEqualizeHistogram.h b/arm_compute/runtime/CL/functions/CLEqualizeHistogram.h
index 883f330b33..17352d1a9b 100644
--- a/arm_compute/runtime/CL/functions/CLEqualizeHistogram.h
+++ b/arm_compute/runtime/CL/functions/CLEqualizeHistogram.h
@@ -24,16 +24,19 @@
 #ifndef ARM_COMPUTE_CLEQUALIZEHISTOGRAM_H
 #define ARM_COMPUTE_CLEQUALIZEHISTOGRAM_H
 
-#include "arm_compute/core/CL/kernels/CLHistogramKernel.h"
-#include "arm_compute/core/CL/kernels/CLTableLookupKernel.h"
 #include "arm_compute/runtime/CL/CLDistribution1D.h"
 #include "arm_compute/runtime/CL/CLLut.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLHistogramKernel;
+class CLHistogramBorderKernel;
+class CLTableLookupKernel;
 class ICLTensor;
 using ICLImage = ICLTensor;
 
@@ -48,6 +51,12 @@ class CLEqualizeHistogram : public IFunction
 public:
     /** Default Constructor. */
     CLEqualizeHistogram();
+    /** Prevent instances of this class from being copied */
+    CLEqualizeHistogram(const CLEqualizeHistogram &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLEqualizeHistogram &operator=(const CLEqualizeHistogram &) = delete;
+    /** Default destructor */
+    ~CLEqualizeHistogram();
     /** Initialise the kernel's inputs.
      *
      * @param[in]  input  Input image. Data types supported: U8.
@@ -66,14 +75,14 @@ public:
     void run() override;
 
 private:
-    CLHistogramKernel       _histogram_kernel;        /**< Kernel that calculates the histogram of input. */
-    CLHistogramBorderKernel _border_histogram_kernel; /**< Kernel that calculates the histogram on the borders. */
-    CLTableLookupKernel     _map_histogram_kernel;    /**< Kernel that maps the input to output using the lut. */
-    CLDistribution1D        _hist;                    /**< Distribution that holds the histogram of the input image. */
-    CLDistribution1D        _cum_dist;                /**< Distribution that holds the cummulative distribution of the input histogram. */
-    CLLut                   _cd_lut;                  /**< Holds the equalization lookuptable. */
-    static const uint32_t   max_range = 256;          /**< Histogram range of the internal histograms. */
-    static const uint32_t   nr_bins   = 256;          /**< Histogram bins of the internal histograms. */
+    std::unique_ptr<CLHistogramKernel>       _histogram_kernel;        /**< Kernel that calculates the histogram of input. */
+    std::unique_ptr<CLHistogramBorderKernel> _border_histogram_kernel; /**< Kernel that calculates the histogram on the borders. */
+    std::unique_ptr<CLTableLookupKernel>     _map_histogram_kernel;    /**< Kernel that maps the input to output using the lut. */
+    CLDistribution1D                         _hist;                    /**< Distribution that holds the histogram of the input image. */
+    CLDistribution1D                         _cum_dist;                /**< Distribution that holds the cummulative distribution of the input histogram. */
+    CLLut                                    _cd_lut;                  /**< Holds the equalization lookuptable. */
+    static const uint32_t                    max_range = 256;          /**< Histogram range of the internal histograms. */
+    static const uint32_t                    nr_bins   = 256;          /**< Histogram bins of the internal histograms. */
 };
 }
 #endif /*ARM_COMPUTE_CLEQUALIZEHISTOGRAM_H */
diff --git a/arm_compute/runtime/CL/functions/CLErode.h b/arm_compute/runtime/CL/functions/CLErode.h
index f8f1c72bc0..9d799bc91e 100644
--- a/arm_compute/runtime/CL/functions/CLErode.h
+++ b/arm_compute/runtime/CL/functions/CLErode.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute erode. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLFFT1D.h b/arm_compute/runtime/CL/functions/CLFFT1D.h
index a6a35ab320..31a2cc6b06 100644
--- a/arm_compute/runtime/CL/functions/CLFFT1D.h
+++ b/arm_compute/runtime/CL/functions/CLFFT1D.h
@@ -26,9 +26,6 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLFFTDigitReverseKernel.h"
-#include "arm_compute/core/CL/kernels/CLFFTRadixStageKernel.h"
-#include "arm_compute/core/CL/kernels/CLFFTScaleKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/FunctionDescriptors.h"
 #include "arm_compute/runtime/MemoryGroup.h"
@@ -36,6 +33,9 @@
 namespace arm_compute
 {
 // Forward declaration
+class CLFFTDigitReverseKernel;
+class CLFFTRadixStageKernel;
+class CLFFTScaleKernel;
 class ICLTensor;
 
 /** Basic function to execute one dimensional FFT. This function calls the following OpenCL kernels:
@@ -49,6 +49,12 @@ class CLFFT1D : public IFunction
 public:
     /** Default Constructor */
     CLFFT1D(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLFFT1D(const CLFFT1D &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLFFT1D &operator=(const CLFFT1D &) = delete;
+    /** Default destructor */
+    ~CLFFT1D();
     /** Initialise the function's source, destinations and border mode.
      *
      * @param[in]  input  Source tensor. Data types supported: F32.
@@ -78,14 +84,14 @@ public:
     void run() override;
 
 protected:
-    MemoryGroup                        _memory_group;
-    CLFFTDigitReverseKernel            _digit_reverse_kernel;
-    std::vector<CLFFTRadixStageKernel> _fft_kernels;
-    CLFFTScaleKernel                   _scale_kernel;
-    CLTensor                           _digit_reversed_input;
-    CLTensor                           _digit_reverse_indices;
-    unsigned int                       _num_ffts;
-    bool                               _run_scale;
+    MemoryGroup                                         _memory_group;
+    std::unique_ptr<CLFFTDigitReverseKernel>            _digit_reverse_kernel;
+    std::vector<std::unique_ptr<CLFFTRadixStageKernel>> _fft_kernels;
+    std::unique_ptr<CLFFTScaleKernel>                   _scale_kernel;
+    CLTensor                                            _digit_reversed_input;
+    CLTensor                                            _digit_reverse_indices;
+    unsigned int                                        _num_ffts;
+    bool                                                _run_scale;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLFFT1D_H */
diff --git a/arm_compute/runtime/CL/functions/CLFFT2D.h b/arm_compute/runtime/CL/functions/CLFFT2D.h
index 9ceebeaa32..126944b323 100644
--- a/arm_compute/runtime/CL/functions/CLFFT2D.h
+++ b/arm_compute/runtime/CL/functions/CLFFT2D.h
@@ -46,6 +46,12 @@ class CLFFT2D : public IFunction
 public:
     /** Default Constructor */
     CLFFT2D(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLFFT2D(const CLFFT2D &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLFFT2D &operator=(const CLFFT2D &) = delete;
+    /** Default destructor */
+    ~CLFFT2D();
     /** Initialise the function's source, destinations and border mode.
      *
      * @param[in]  input  Source tensor. Data types supported: F32.
diff --git a/arm_compute/runtime/CL/functions/CLFastCorners.h b/arm_compute/runtime/CL/functions/CLFastCorners.h
index 698cc67995..e110582c50 100644
--- a/arm_compute/runtime/CL/functions/CLFastCorners.h
+++ b/arm_compute/runtime/CL/functions/CLFastCorners.h
@@ -25,7 +25,6 @@
 #define ARM_COMPUTE_CLFASTCORNERS_H
 
 #include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/CL/kernels/CLFastCornersKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Window.h"
 #include "arm_compute/runtime/CL/CLArray.h"
@@ -40,6 +39,8 @@
 
 namespace arm_compute
 {
+class CLFastCornersKernel;
+class CLCopyToArrayKernel;
 class ICLTensor;
 using ICLImage = ICLTensor;
 
@@ -59,6 +60,8 @@ public:
     CLFastCorners(const CLFastCorners &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     const CLFastCorners &operator=(const CLFastCorners &) = delete;
+    /** Default destructor */
+    ~CLFastCorners();
     /** Initialize the function's source, destination, conv and border_mode.
      *
      * @param[in]     input                 Source image. Data types supported: U8.
@@ -88,18 +91,18 @@ public:
     void run() override;
 
 private:
-    MemoryGroup               _memory_group;
-    CLFastCornersKernel       _fast_corners_kernel;
-    CLNonMaximaSuppression3x3 _suppr_func;
-    CLCopyToArrayKernel       _copy_array_kernel;
-    CLImage                   _output;
-    CLImage                   _suppr;
-    Window                    _win;
-    bool                      _non_max;
-    unsigned int             *_num_corners;
-    cl::Buffer                _num_buffer;
-    ICLKeyPointArray         *_corners;
-    uint8_t                   _constant_border_value;
+    MemoryGroup                          _memory_group;
+    std::unique_ptr<CLFastCornersKernel> _fast_corners_kernel;
+    CLNonMaximaSuppression3x3            _suppr_func;
+    std::unique_ptr<CLCopyToArrayKernel> _copy_array_kernel;
+    CLImage                              _output;
+    CLImage                              _suppr;
+    Window                               _win;
+    bool                                 _non_max;
+    unsigned int                        *_num_corners;
+    cl::Buffer                           _num_buffer;
+    ICLKeyPointArray                    *_corners;
+    uint8_t                              _constant_border_value;
 };
 }
 #endif /*ARM_COMPUTE_CLFASTCORNERS_H */
diff --git a/arm_compute/runtime/CL/functions/CLFill.h b/arm_compute/runtime/CL/functions/CLFill.h
index b79b234158..fef8324432 100644
--- a/arm_compute/runtime/CL/functions/CLFill.h
+++ b/arm_compute/runtime/CL/functions/CLFill.h
@@ -30,6 +30,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Function to run @ref CLMemsetKernel to fill a tensor with a scalar value */
diff --git a/arm_compute/runtime/CL/functions/CLFillBorder.h b/arm_compute/runtime/CL/functions/CLFillBorder.h
index 18bc20e654..a4ad82dfd4 100644
--- a/arm_compute/runtime/CL/functions/CLFillBorder.h
+++ b/arm_compute/runtime/CL/functions/CLFillBorder.h
@@ -30,6 +30,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLFillBorderKernel */
diff --git a/arm_compute/runtime/CL/functions/CLFlattenLayer.h b/arm_compute/runtime/CL/functions/CLFlattenLayer.h
index b8139c2260..f5f4ff554f 100644
--- a/arm_compute/runtime/CL/functions/CLFlattenLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFlattenLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute flatten. This function calls the following OpenCL kernel:
 *
diff --git a/arm_compute/runtime/CL/functions/CLFloor.h b/arm_compute/runtime/CL/functions/CLFloor.h
index 93c3639f89..85d7071194 100644
--- a/arm_compute/runtime/CL/functions/CLFloor.h
+++ b/arm_compute/runtime/CL/functions/CLFloor.h
@@ -30,7 +30,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLFloorKernel */
 class CLFloor : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
index 29788742d7..3f17e4a921 100644
--- a/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h
@@ -26,7 +26,6 @@
 
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLTransposeKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLConvertFullyConnectedWeights.h"
 #include "arm_compute/runtime/CL/functions/CLFlattenLayer.h"
diff --git a/arm_compute/runtime/CL/functions/CLFuseBatchNormalization.h b/arm_compute/runtime/CL/functions/CLFuseBatchNormalization.h
index de6d5617c2..e35905fcf1 100644
--- a/arm_compute/runtime/CL/functions/CLFuseBatchNormalization.h
+++ b/arm_compute/runtime/CL/functions/CLFuseBatchNormalization.h
@@ -24,14 +24,18 @@
 #ifndef ARM_COMPUTE_CLFUSEBATCHNORMALIZATION_H
 #define ARM_COMPUTE_CLFUSEBATCHNORMALIZATION_H
 
-#include "arm_compute/core/CL/kernels/CLFuseBatchNormalizationKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
+class CLFuseBatchNormalizationKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to fuse the batch normalization node to a preceding convolution node */
 class CLFuseBatchNormalization : public IFunction
@@ -48,7 +52,7 @@ public:
     /** Allow instances of this class to be moved */
     CLFuseBatchNormalization &operator=(CLFuseBatchNormalization &&) = default;
     /** Default destructor */
-    ~CLFuseBatchNormalization() = default;
+    ~CLFuseBatchNormalization();
     /** Set the input and output tensors.
      *
      * @param[in]  input_weights Input weights tensor for convolution or depthwise convolution layer. Data type supported: F16/F32. Data layout supported: NCHW, NHWC
@@ -112,7 +116,7 @@ public:
     void run() override;
 
 private:
-    CLFuseBatchNormalizationKernel _fuse_bn_kernel;
+    std::unique_ptr<CLFuseBatchNormalizationKernel> _fuse_bn_kernel;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLFUSEBATCHNORMALIZATION_H */
diff --git a/arm_compute/runtime/CL/functions/CLGEMM.h b/arm_compute/runtime/CL/functions/CLGEMM.h
index 92f9736e35..0b13e7dbbf 100644
--- a/arm_compute/runtime/CL/functions/CLGEMM.h
+++ b/arm_compute/runtime/CL/functions/CLGEMM.h
@@ -24,11 +24,6 @@
 #ifndef ARM_COMPUTE_CLGEMM_H
 #define ARM_COMPUTE_CLGEMM_H
 
-#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/CLTypes.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -36,9 +31,18 @@
 #include "arm_compute/runtime/IWeightsManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLGEMMReshapeRHSMatrixKernel;
+class CLGEMMMatrixMultiplyKernel;
+class CLGEMMMatrixMultiplyReshapedKernel;
+class CLGEMMMatrixMultiplyReshapedOnlyRHSKernel;
+class CLGEMMReshapeLHSMatrixKernel;
 class ICLTensor;
+class ITensorInfo;
 
 namespace weights_transformations
 {
@@ -46,41 +50,36 @@ namespace weights_transformations
 class CLGEMMReshapeRHSMatrixKernelManaged : public ITransformWeights
 {
 public:
+    /** Default constructor */
+    CLGEMMReshapeRHSMatrixKernelManaged();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMReshapeRHSMatrixKernelManaged(const CLGEMMReshapeRHSMatrixKernelManaged &) = delete;
+    /** Default move constructor */
+    CLGEMMReshapeRHSMatrixKernelManaged(CLGEMMReshapeRHSMatrixKernelManaged &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMReshapeRHSMatrixKernelManaged &operator=(const CLGEMMReshapeRHSMatrixKernelManaged &) = delete;
+    /** Default move assignment operator */
+    CLGEMMReshapeRHSMatrixKernelManaged &operator=(CLGEMMReshapeRHSMatrixKernelManaged &&) = default;
+    /** Default desctructor */
+    ~CLGEMMReshapeRHSMatrixKernelManaged();
     //Inherited method override
-    void run() override
-    {
-        _output.allocator()->allocate();
-        CLScheduler::get().enqueue(_kernel, false);
-        _reshape_run = true;
-    }
+    void run() override;
 
     //Inherited method override
-    void release() override
-    {
-        _output.allocator()->free();
-    }
+    void release() override;
 
     //Inherited method override
-    ICLTensor *get_weights() override
-    {
-        return &_output;
-    }
+    ICLTensor *get_weights() override;
 
     //Inherited method override
-    uint32_t uid() override
-    {
-        return _uid;
-    }
+    uint32_t uid() override;
 
     /** Configures the @ref CLGEMMReshapeRHSMatrixKernel kernel
      *
      * @param[in] input Input tensor. Data types supported: All
      * @param[in] info  RHS matrix information to be used for reshaping.
      */
-    void configure(const ICLTensor *input, GEMMRHSMatrixInfo info)
-    {
-        configure(CLKernelLibrary::get().get_compile_context(), input, info);
-    }
+    void configure(const ICLTensor *input, GEMMRHSMatrixInfo info);
 
     /** Configures the @ref CLGEMMReshapeRHSMatrixKernel kernel
      *
@@ -88,15 +87,12 @@ public:
      * @param[in] input           Input tensor. Data types supported: All
      * @param[in] info            RHS matrix information to be used for reshaping.
      */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, GEMMRHSMatrixInfo info)
-    {
-        _kernel.configure(compile_context, input, &_output, info);
-    }
+    void configure(const CLCompileContext &compile_context, const ICLTensor *input, GEMMRHSMatrixInfo info);
 
 private:
-    static constexpr uint32_t    _uid = 0x15;
-    CLTensor                     _output{};
-    CLGEMMReshapeRHSMatrixKernel _kernel{};
+    static constexpr uint32_t                     _uid{ 0x15 };
+    CLTensor                                      _output{};
+    std::unique_ptr<CLGEMMReshapeRHSMatrixKernel> _kernel;
 };
 } // namespace weights_transformations
 
@@ -126,6 +122,8 @@ public:
     CLGEMM &operator=(const CLGEMM &) = delete;
     /** Default move assignment operator */
     CLGEMM &operator=(CLGEMM &&) = default;
+    /** Default destructor */
+    ~CLGEMM();
     /** Initialise the kernel's inputs and output
      *
      * @note GEMM: General Matrix Multiply - [alpha * A * B + beta * C].
@@ -198,24 +196,24 @@ private:
     static Status validate_reshaped(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, float alpha, float beta, const GEMMInfo &gemm_info);
     static Status validate_reshaped_only_rhs(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, float alpha, float beta, const GEMMInfo &gemm_info);
 
-    MemoryGroup                                                  _memory_group;
-    IWeightsManager                                             *_weights_manager;
-    CLGEMMMatrixMultiplyKernel                                   _mm_kernel;
-    CLGEMMReshapeLHSMatrixKernel                                 _reshape_lhs_kernel;
-    CLGEMMReshapeRHSMatrixKernel                                 _reshape_rhs_kernel;
-    weights_transformations::CLGEMMReshapeRHSMatrixKernelManaged _reshape_rhs_kernel_managed;
-    CLGEMMMatrixMultiplyReshapedKernel                           _mm_reshaped_kernel;
-    CLGEMMMatrixMultiplyReshapedOnlyRHSKernel                    _mm_reshaped_only_rhs_kernel;
-    CLGEMMMatrixMultiplyReshapedOnlyRHSKernel                    _mm_reshaped_only_rhs_fallback_kernel;
-    CLTensor                                                     _tmp_a;
-    CLTensor                                                     _tmp_b;
-    const ICLTensor                                             *_original_b;
-    const ICLTensor                                             *_lhs;
-    ICLTensor                                                   *_dst;
-    bool                                                         _reshape_b_only_on_first_run;
-    bool                                                         _is_prepared;
-    bool                                                         _has_pad_y;
-    CLGEMMKernelType                                             _gemm_kernel_type;
+    MemoryGroup                                                                   _memory_group;
+    IWeightsManager                                                              *_weights_manager;
+    std::unique_ptr<CLGEMMMatrixMultiplyKernel>                                   _mm_kernel;
+    std::unique_ptr<CLGEMMReshapeLHSMatrixKernel>                                 _reshape_lhs_kernel;
+    std::unique_ptr<CLGEMMReshapeRHSMatrixKernel>                                 _reshape_rhs_kernel;
+    std::unique_ptr<weights_transformations::CLGEMMReshapeRHSMatrixKernelManaged> _reshape_rhs_kernel_managed;
+    std::unique_ptr<CLGEMMMatrixMultiplyReshapedKernel>                           _mm_reshaped_kernel;
+    std::unique_ptr<CLGEMMMatrixMultiplyReshapedOnlyRHSKernel>                    _mm_reshaped_only_rhs_kernel;
+    std::unique_ptr<CLGEMMMatrixMultiplyReshapedOnlyRHSKernel>                    _mm_reshaped_only_rhs_fallback_kernel;
+    CLTensor                                                                      _tmp_a;
+    CLTensor                                                                      _tmp_b;
+    const ICLTensor                                                              *_original_b;
+    const ICLTensor                                                              *_lhs;
+    ICLTensor                                                                    *_dst;
+    bool                                                                          _reshape_b_only_on_first_run;
+    bool                                                                          _is_prepared;
+    bool                                                                          _has_pad_y;
+    CLGEMMKernelType                                                              _gemm_kernel_type;
 };
 } // namespace arm_compute
 
diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
index 467045cd86..340ac6e749 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
@@ -26,9 +26,7 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLCol2ImKernel.h"
-#include "arm_compute/core/CL/kernels/CLIm2ColKernel.h"
-#include "arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
@@ -43,6 +41,9 @@
 
 namespace arm_compute
 {
+class CLCol2ImKernel;
+class CLIm2ColKernel;
+class CLWeightsReshapeKernel;
 class ICLTensor;
 
 /** Function to reshape and transpose the weights. This function calls the following kernels:
@@ -53,6 +54,16 @@ class CLConvolutionLayerReshapeWeights : public IFunction
 public:
     /** Constructor */
     CLConvolutionLayerReshapeWeights();
+    /** Prevent instances of this class from being copied */
+    CLConvolutionLayerReshapeWeights(const CLConvolutionLayerReshapeWeights &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLConvolutionLayerReshapeWeights &operator=(const CLConvolutionLayerReshapeWeights &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLConvolutionLayerReshapeWeights(CLConvolutionLayerReshapeWeights &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLConvolutionLayerReshapeWeights &operator=(CLConvolutionLayerReshapeWeights &&) = delete;
+    /** Default destructor */
+    ~CLConvolutionLayerReshapeWeights();
     /** Set the input and output tensors.
      *
      * @param[in]  weights    Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
@@ -87,7 +98,7 @@ public:
     void run() override;
 
 private:
-    CLWeightsReshapeKernel _weights_reshape_kernel;
+    std::unique_ptr<CLWeightsReshapeKernel> _weights_reshape_kernel;
 };
 
 namespace weights_transformations
@@ -179,6 +190,8 @@ public:
     CLGEMMConvolutionLayer &operator=(const CLGEMMConvolutionLayer &) = delete;
     /** Default move assignment operator */
     CLGEMMConvolutionLayer &operator=(CLGEMMConvolutionLayer &&) = default;
+    /**Default destructor */
+    ~CLGEMMConvolutionLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input        Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
@@ -288,10 +301,10 @@ private:
     IWeightsManager                                                   *_weights_manager;
     CLConvolutionLayerReshapeWeights                                   _reshape_weights;
     weights_transformations::CLConvolutionLayerReshapeWeightsTransform _reshape_weights_managed;
-    CLIm2ColKernel                                                     _im2col_kernel;
+    std::unique_ptr<CLIm2ColKernel>                                    _im2col_kernel;
     CLGEMM                                                             _mm_gemm;
     CLGEMMLowpMatrixMultiplyCore                                       _mm_gemmlowp;
-    CLCol2ImKernel                                                     _col2im_kernel;
+    std::unique_ptr<CLCol2ImKernel>                                    _col2im_kernel;
     CLActivationLayer                                                  _activationlayer_function;
 
     const ICLTensor *_original_weights;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMDeconvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMDeconvolutionLayer.h
index 1fedeff444..32af0f9427 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMDeconvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMDeconvolutionLayer.h
@@ -24,7 +24,6 @@
 #ifndef ARM_COMPUTE_CLGEMMDECONVOLUTIONLAYER_H
 #define ARM_COMPUTE_CLGEMMDECONVOLUTIONLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLConvolutionLayer.h"
 #include "arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h"
@@ -40,6 +39,7 @@
 
 namespace arm_compute
 {
+class CLDeconvolutionReshapeOutputKernel;
 class ICLTensor;
 /** Function to run the deconvolution layer through a call to GEMM.
  *
@@ -89,6 +89,8 @@ public:
     CLGEMMDeconvolutionLayer &operator=(const CLGEMMDeconvolutionLayer &) = delete;
     /** Default move assignment operator */
     CLGEMMDeconvolutionLayer &operator=(CLGEMMDeconvolutionLayer &&) = default;
+    /** Default desctructor */
+    ~CLGEMMDeconvolutionLayer();
     /** Set the input, weights, biases and output tensors.
      *
      * @param[in,out] input       Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs.
@@ -130,15 +132,15 @@ public:
 private:
     MemoryGroup _memory_group;
 
-    CLGEMM                             _mm_gemm;
-    CLGEMMLowpMatrixMultiplyCore       _mm_gemmlowp;
-    CLGEMMLowpOutputStage              _gemmlowp_output_stage;
-    CLPermute                          _permute_input_to_nhwc;
-    CLPermute                          _permute_weights_to_nhwc;
-    CLReshapeLayer                     _reshape_weights;
-    CLTranspose                        _transpose_weights;
-    CLDeconvolutionReshapeOutputKernel _deconv_reshape;
-    CLSlice                            _slice_gemm;
+    CLGEMM                                              _mm_gemm;
+    CLGEMMLowpMatrixMultiplyCore                        _mm_gemmlowp;
+    CLGEMMLowpOutputStage                               _gemmlowp_output_stage;
+    CLPermute                                           _permute_input_to_nhwc;
+    CLPermute                                           _permute_weights_to_nhwc;
+    CLReshapeLayer                                      _reshape_weights;
+    CLTranspose                                         _transpose_weights;
+    std::unique_ptr<CLDeconvolutionReshapeOutputKernel> _deconv_reshape;
+    CLSlice                                             _slice_gemm;
 
     CLTensor _gemmlowp_final;
     CLTensor _reshaped_weights;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
index 57b1e30df5..4cc8899690 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
@@ -24,21 +24,24 @@
 #ifndef ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H
 #define ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H
 
-#include "arm_compute/core/CL/kernels/CLDepthConvertLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionOutputStageKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/MemoryGroup.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class IMemoryManager;
 class ICLTensor;
+class ITensorInfo;
+class CLDepthConvertLayerKernel;
+class CLGEMMLowpMatrixMultiplyNativeKernel;
+class CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel;
+class CLGEMMLowpOffsetContributionKernel;
+class CLGEMMLowpOffsetContributionOutputStageKernel;
+class CLGEMMLowpMatrixAReductionKernel;
+class CLGEMMLowpMatrixBReductionKernel;
+class CLGEMMReshapeRHSMatrixKernel;
 
 /** Basic function to execute GEMMLowpMatrixMultiplyCore on OpenCL. */
 class CLGEMMLowpMatrixMultiplyCore : public IFunction
@@ -54,6 +57,8 @@ public:
     CLGEMMLowpMatrixMultiplyCore &operator=(const CLGEMMLowpMatrixMultiplyCore &) = delete;
     /** Default move assignment operator */
     CLGEMMLowpMatrixMultiplyCore &operator=(CLGEMMLowpMatrixMultiplyCore &&) = default;
+    /** Default destructor */
+    ~CLGEMMLowpMatrixMultiplyCore();
     /** Initialise the kernel's inputs, output
      *
      * @note GEMMLowp:  low precision GEMM kernel. [A * B + C]
@@ -112,14 +117,14 @@ private:
     MemoryGroup _memory_group;
 
     // Kernels used
-    CLDepthConvertLayerKernel                     _weights_to_qasymm8;
-    CLGEMMLowpMatrixMultiplyNativeKernel          _mm_native_kernel;
-    CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel _mm_reshaped_only_rhs_kernel;
-    CLGEMMReshapeRHSMatrixKernel                  _mtx_b_reshape_kernel;
-    CLGEMMLowpMatrixAReductionKernel              _mtx_a_reduction_kernel;
-    CLGEMMLowpMatrixBReductionKernel              _mtx_b_reduction_kernel;
-    CLGEMMLowpOffsetContributionKernel            _offset_contribution_kernel;
-    CLGEMMLowpOffsetContributionOutputStageKernel _offset_contribution_output_stage_kernel;
+    std::unique_ptr<CLDepthConvertLayerKernel>                     _weights_to_qasymm8;
+    std::unique_ptr<CLGEMMLowpMatrixMultiplyNativeKernel>          _mm_native_kernel;
+    std::unique_ptr<CLGEMMLowpMatrixMultiplyReshapedOnlyRHSKernel> _mm_reshaped_only_rhs_kernel;
+    std::unique_ptr<CLGEMMReshapeRHSMatrixKernel>                  _mtx_b_reshape_kernel;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel>              _mtx_a_reduction_kernel;
+    std::unique_ptr<CLGEMMLowpMatrixBReductionKernel>              _mtx_b_reduction_kernel;
+    std::unique_ptr<CLGEMMLowpOffsetContributionKernel>            _offset_contribution_kernel;
+    std::unique_ptr<CLGEMMLowpOffsetContributionOutputStageKernel> _offset_contribution_output_stage_kernel;
 
     // Temporary tensors
     CLTensor _qasymm8_weights;
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
index 44c52ffb79..a4edab9b8f 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h
@@ -24,8 +24,11 @@
 #ifndef ARM_COMPUTE_CLGEMMLOWPOUTPUTSTAGE_H
 #define ARM_COMPUTE_CLGEMMLOWPOUTPUTSTAGE_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
+#include <limits>
+
 /** This file contains all available output stages for GEMMLowp on OpenCL.
  *
  *  In gemmlowp, the "output stage" is the process that takes a final int32 accumulator value (the output of @ref CLGEMMLowpMatrixMultiplyCore),
@@ -36,7 +39,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ITensor;
+class ICLTensor;
+class ITensorInfo;
+struct GEMMLowpOutputStageInfo;
 
 /** Basic function to execute CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPoint on OpenCL.
  *
diff --git a/arm_compute/runtime/CL/functions/CLGather.h b/arm_compute/runtime/CL/functions/CLGather.h
index e87a120ba1..9c659be6fc 100644
--- a/arm_compute/runtime/CL/functions/CLGather.h
+++ b/arm_compute/runtime/CL/functions/CLGather.h
@@ -25,11 +25,14 @@
 #ifndef ARM_COMPUTE_CLGATHER_H
 #define ARM_COMPUTE_CLGATHER_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLGatherKernel */
 class CLGather : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLGaussian3x3.h b/arm_compute/runtime/CL/functions/CLGaussian3x3.h
index 9fe3e9bb00..286a17618b 100644
--- a/arm_compute/runtime/CL/functions/CLGaussian3x3.h
+++ b/arm_compute/runtime/CL/functions/CLGaussian3x3.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute gaussian filter 3x3. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLGaussian5x5.h b/arm_compute/runtime/CL/functions/CLGaussian5x5.h
index fb369d750b..cf5b79eaac 100644
--- a/arm_compute/runtime/CL/functions/CLGaussian5x5.h
+++ b/arm_compute/runtime/CL/functions/CLGaussian5x5.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLGAUSSIAN5X5_H
 #define ARM_COMPUTE_CLGAUSSIAN5X5_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLGaussian5x5Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -37,6 +35,10 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLGaussian5x5HorKernel;
+class CLGaussian5x5VertKernel;
 class ICLTensor;
 
 /** Basic function to execute gaussian filter 5x5. This function calls the following OpenCL kernels:
@@ -54,6 +56,16 @@ public:
      * @param[in] memory_manager (Optional) Memory manager.
      */
     CLGaussian5x5(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLGaussian5x5(const CLGaussian5x5 &) = delete;
+    /** Default move constructor */
+    CLGaussian5x5(CLGaussian5x5 &&) = default;
+    /** Prevent instances of this class from being copied */
+    CLGaussian5x5 &operator=(const CLGaussian5x5 &) = delete;
+    /** Default move assignment operator */
+    CLGaussian5x5 &operator=(CLGaussian5x5 &&) = default;
+    /** Default destructor */
+    ~CLGaussian5x5();
     /** Initialise the function's source, destinations and border mode.
      *
      * @param[in,out] input                 Source tensor. Data types supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -76,11 +88,11 @@ public:
     void run() override;
 
 protected:
-    MemoryGroup             _memory_group;   /**< Function's memory group */
-    CLGaussian5x5HorKernel  _kernel_hor;     /**< Horizontal pass kernel */
-    CLGaussian5x5VertKernel _kernel_vert;    /**< Vertical pass kernel */
-    CLFillBorderKernel      _border_handler; /**< Kernel to handle image borders */
-    CLImage                 _tmp;            /**< Temporary buffer */
+    MemoryGroup                              _memory_group;   /**< Function's memory group */
+    std::unique_ptr<CLGaussian5x5HorKernel>  _kernel_hor;     /**< Horizontal pass kernel */
+    std::unique_ptr<CLGaussian5x5VertKernel> _kernel_vert;    /**< Vertical pass kernel */
+    std::unique_ptr<CLFillBorderKernel>      _border_handler; /**< Kernel to handle image borders */
+    CLImage                                  _tmp;            /**< Temporary buffer */
 };
 }
 #endif /*ARM_COMPUTE_CLGAUSSIAN5X5_H */
diff --git a/arm_compute/runtime/CL/functions/CLGaussianPyramid.h b/arm_compute/runtime/CL/functions/CLGaussianPyramid.h
index 70f324be11..b18e5f98f0 100644
--- a/arm_compute/runtime/CL/functions/CLGaussianPyramid.h
+++ b/arm_compute/runtime/CL/functions/CLGaussianPyramid.h
@@ -24,9 +24,6 @@
 #ifndef ARM_COMPUTE_CLGAUSSIANPYRAMID_H
 #define ARM_COMPUTE_CLGAUSSIANPYRAMID_H
 
-#include "arm_compute/core/CL/kernels/CLGaussianPyramidKernel.h"
-
-#include "arm_compute/core/CL/kernels/CLScaleKernel.h"
 #include "arm_compute/core/IPyramid.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLPyramid.h"
@@ -38,7 +35,12 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
 class ICLTensor;
+class CLGaussianPyramidHorKernel;
+class CLGaussianPyramidVertKernel;
+class CLScaleKernel;
 
 /** Common interface for all Gaussian pyramid functions */
 class CLGaussianPyramid : public IFunction
@@ -55,7 +57,7 @@ public:
     /** Allow instances of this class to be moved */
     CLGaussianPyramid &operator=(CLGaussianPyramid &&) = default;
     /** Default destructor */
-    virtual ~CLGaussianPyramid() = default;
+    ~CLGaussianPyramid();
     /** Initialise the function's source, destinations and border mode.
      *
      * @param[in, out] input                 Source tensor. Data types supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -93,6 +95,12 @@ class CLGaussianPyramidHalf : public CLGaussianPyramid
 public:
     /** Constructor */
     CLGaussianPyramidHalf();
+    /** Prevent instances of this class from being copied */
+    CLGaussianPyramidHalf(const CLGaussianPyramidHalf &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLGaussianPyramidHalf &operator=(const CLGaussianPyramidHalf &) = delete;
+    /** Default destructor */
+    ~CLGaussianPyramidHalf();
 
     // Inherited methods overridden:
     void configure(ICLTensor *input, CLPyramid *pyramid, BorderMode border_mode, uint8_t constant_border_value) override;
@@ -100,10 +108,10 @@ public:
     void run() override;
 
 private:
-    std::vector<CLFillBorderKernel>          _horizontal_border_handler;
-    std::vector<CLFillBorderKernel>          _vertical_border_handler;
-    std::vector<CLGaussianPyramidHorKernel>  _horizontal_reduction;
-    std::vector<CLGaussianPyramidVertKernel> _vertical_reduction;
+    std::vector<std::unique_ptr<CLFillBorderKernel>>          _horizontal_border_handler;
+    std::vector<std::unique_ptr<CLFillBorderKernel>>          _vertical_border_handler;
+    std::vector<std::unique_ptr<CLGaussianPyramidHorKernel>>  _horizontal_reduction;
+    std::vector<std::unique_ptr<CLGaussianPyramidVertKernel>> _vertical_reduction;
 };
 
 /** Basic function to execute gaussian pyramid with ORB scale factor. This function calls the following OpenCL kernels and functions:
@@ -124,8 +132,8 @@ public:
     void run() override;
 
 private:
-    std::vector<CLGaussian5x5> _gauss5x5;
-    std::vector<CLScaleKernel> _scale_nearest;
+    std::vector<CLGaussian5x5>                  _gauss5x5;
+    std::vector<std::unique_ptr<CLScaleKernel>> _scale_nearest;
 };
 }
 #endif /*ARM_COMPUTE_CLGAUSSIANPYRAMID_H */
diff --git a/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h b/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
index 6d5f2e5d71..0fb9a06c84 100644
--- a/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
@@ -23,12 +23,7 @@
  */
 #ifndef ARM_COMPUTE_CLGENERATEPROPOSALSLAYER_H
 #define ARM_COMPUTE_CLGENERATEPROPOSALSLAYER_H
-#include "arm_compute/core/CL/kernels/CLBoundingBoxTransformKernel.h"
-#include "arm_compute/core/CL/kernels/CLDequantizationLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLGenerateProposalsLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLPadLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLPermuteKernel.h"
-#include "arm_compute/core/CL/kernels/CLQuantizationLayerKernel.h"
+
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
@@ -38,9 +33,19 @@
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/MemoryGroup.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLBoundingBoxTransformKernel;
+class CLDequantizationLayerKernel;
+class CLComputeAllAnchorsKernel;
+class CLPadLayerKernel;
+class CLPermuteKernel;
+class CLQuantizationLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to generate proposals for a RPN (Region Proposal Network)
  *
@@ -67,6 +72,8 @@ public:
     CLGenerateProposalsLayer(const CLGenerateProposalsLayer &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLGenerateProposalsLayer &operator=(const CLGenerateProposalsLayer &) = delete;
+    /** Default destructor */
+    ~CLGenerateProposalsLayer();
 
     /** Set the input and output tensors.
      *
@@ -130,16 +137,16 @@ private:
     MemoryGroup _memory_group;
 
     // OpenCL kernels
-    CLPermuteKernel              _permute_deltas_kernel;
-    CLReshapeLayer               _flatten_deltas;
-    CLPermuteKernel              _permute_scores_kernel;
-    CLReshapeLayer               _flatten_scores;
-    CLComputeAllAnchorsKernel    _compute_anchors_kernel;
-    CLBoundingBoxTransformKernel _bounding_box_kernel;
-    CLPadLayerKernel             _pad_kernel;
-    CLDequantizationLayerKernel  _dequantize_anchors;
-    CLDequantizationLayerKernel  _dequantize_deltas;
-    CLQuantizationLayerKernel    _quantize_all_proposals;
+    std::unique_ptr<CLPermuteKernel>              _permute_deltas_kernel;
+    CLReshapeLayer                                _flatten_deltas;
+    std::unique_ptr<CLPermuteKernel>              _permute_scores_kernel;
+    CLReshapeLayer                                _flatten_scores;
+    std::unique_ptr<CLComputeAllAnchorsKernel>    _compute_anchors_kernel;
+    std::unique_ptr<CLBoundingBoxTransformKernel> _bounding_box_kernel;
+    std::unique_ptr<CLPadLayerKernel>             _pad_kernel;
+    std::unique_ptr<CLDequantizationLayerKernel>  _dequantize_anchors;
+    std::unique_ptr<CLDequantizationLayerKernel>  _dequantize_deltas;
+    std::unique_ptr<CLQuantizationLayerKernel>    _quantize_all_proposals;
 
     // CPP functions
     CPPBoxWithNonMaximaSuppressionLimit _cpp_nms;
diff --git a/arm_compute/runtime/CL/functions/CLHOGDescriptor.h b/arm_compute/runtime/CL/functions/CLHOGDescriptor.h
index dad7e6edf8..fa37b3c84e 100644
--- a/arm_compute/runtime/CL/functions/CLHOGDescriptor.h
+++ b/arm_compute/runtime/CL/functions/CLHOGDescriptor.h
@@ -24,7 +24,6 @@
 #ifndef ARM_COMPUTE_CLHOGDESCRIPTOR_H
 #define ARM_COMPUTE_CLHOGDESCRIPTOR_H
 
-#include "arm_compute/core/CL/kernels/CLHOGDescriptorKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLHOGGradient.h"
@@ -37,6 +36,8 @@
 namespace arm_compute
 {
 class IHOG;
+class CLHOGOrientationBinningKernel;
+class CLHOGBlockNormalizationKernel;
 /** Basic function to calculate HOG descriptor. This function calls the following OpenCL kernels:
  *
  * -# @ref CLHOGGradient
@@ -49,6 +50,12 @@ class CLHOGDescriptor : public IFunction
 public:
     /** Default constructor */
     CLHOGDescriptor(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLHOGDescriptor(const CLHOGDescriptor &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLHOGDescriptor &operator=(const CLHOGDescriptor &) = delete;
+    /** Default destructor */
+    ~CLHOGDescriptor();
     /** Initialise the function's source, destination, HOG data-object and border mode
      *
      * @param[in, out] input                 Input tensor. Data type supported: U8
@@ -75,13 +82,13 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                   _memory_group;
-    CLHOGGradient                 _gradient;
-    CLHOGOrientationBinningKernel _orient_bin;
-    CLHOGBlockNormalizationKernel _block_norm;
-    CLTensor                      _mag;
-    CLTensor                      _phase;
-    CLTensor                      _hog_space;
+    MemoryGroup                                    _memory_group;
+    CLHOGGradient                                  _gradient;
+    std::unique_ptr<CLHOGOrientationBinningKernel> _orient_bin;
+    std::unique_ptr<CLHOGBlockNormalizationKernel> _block_norm;
+    CLTensor                                       _mag;
+    CLTensor                                       _phase;
+    CLTensor                                       _hog_space;
 };
 }
 
diff --git a/arm_compute/runtime/CL/functions/CLHOGDetector.h b/arm_compute/runtime/CL/functions/CLHOGDetector.h
index 6697b5c24d..edc5b652d3 100644
--- a/arm_compute/runtime/CL/functions/CLHOGDetector.h
+++ b/arm_compute/runtime/CL/functions/CLHOGDetector.h
@@ -24,13 +24,20 @@
 #ifndef ARM_COMPUTE_CLHOGDETECTOR_H
 #define ARM_COMPUTE_CLHOGDETECTOR_H
 
+#include "arm_compute/core/CL/ICLArray.h"
 #include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/CL/kernels/CLHOGDetectorKernel.h"
 #include "arm_compute/core/IHOG.h"
 #include "arm_compute/runtime/IFunction.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLHOGDetectorKernel;
+class ICLTensor;
+class ICLHOG;
+
 /** Basic function to execute HOG detector based on linear SVM. This function calls the following OpenCL kernel:
  *
  * -# @ref CLHOGDetectorKernel
@@ -50,7 +57,7 @@ public:
     /** Allow instances of this class to be moved */
     CLHOGDetector &operator=(CLHOGDetector &&) = default;
     /** Default destructor */
-    ~CLHOGDetector() = default;
+    ~CLHOGDetector();
     /** Initialise the kernel's input, output, HOG data object, detection window stride, threshold and index class
      *
      * @attention The function does not reset the number of values in @ref IDetectionWindowArray so it is caller's responsibility to clear it.
@@ -78,16 +85,16 @@ public:
      * @param[in]  idx_class               (Optional) Index of the class used for evaluating which class the detection window belongs to
      */
     void configure(const CLCompileContext &compile_context, const ICLTensor *input, const ICLHOG *hog, ICLDetectionWindowArray *detection_windows, const Size2D &detection_window_stride,
-                   float threshold = 0.0f,
+                   float  threshold = 0.0f,
                    size_t idx_class = 0);
 
     // Inherited methods overridden:
     void run() override;
 
 private:
-    CLHOGDetectorKernel      _hog_detector_kernel;
-    ICLDetectionWindowArray *_detection_windows;
-    cl::Buffer               _num_detection_windows;
+    std::unique_ptr<CLHOGDetectorKernel> _hog_detector_kernel;
+    ICLDetectionWindowArray             *_detection_windows;
+    cl::Buffer                           _num_detection_windows;
 };
 }
 
diff --git a/arm_compute/runtime/CL/functions/CLHOGGradient.h b/arm_compute/runtime/CL/functions/CLHOGGradient.h
index b0589027e7..39d26fb110 100644
--- a/arm_compute/runtime/CL/functions/CLHOGGradient.h
+++ b/arm_compute/runtime/CL/functions/CLHOGGradient.h
@@ -24,9 +24,6 @@
 #ifndef ARM_COMPUTE_CLHOGGRADIENT_H
 #define ARM_COMPUTE_CLHOGGRADIENT_H
 
-#include "arm_compute/core/CL/ICLKernel.h"
-
-#include "arm_compute/core/CL/kernels/CLMagnitudePhaseKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLDerivative.h"
@@ -39,6 +36,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLMagnitudePhaseKernel;
+class ITensorInfo;
 /** Basic function to calculate the gradient for HOG. This function calls the following OpenCL kernels:
  *
  * -# @ref CLDerivative
@@ -79,11 +79,11 @@ public:
     void run() override;
 
 private:
-    MemoryGroup            _memory_group;
-    CLDerivative           _derivative;
-    CLMagnitudePhaseKernel _mag_phase;
-    CLTensor               _gx;
-    CLTensor               _gy;
+    MemoryGroup                             _memory_group;
+    CLDerivative                            _derivative;
+    std::unique_ptr<CLMagnitudePhaseKernel> _mag_phase;
+    CLTensor                                _gx;
+    CLTensor                                _gy;
 };
 }
 #endif /*ARM_COMPUTE_CLHOGGRADIENT_H */
diff --git a/arm_compute/runtime/CL/functions/CLHOGMultiDetection.h b/arm_compute/runtime/CL/functions/CLHOGMultiDetection.h
index e7631c2c5a..2a2c9a0a5c 100644
--- a/arm_compute/runtime/CL/functions/CLHOGMultiDetection.h
+++ b/arm_compute/runtime/CL/functions/CLHOGMultiDetection.h
@@ -26,7 +26,6 @@
 
 #include "arm_compute/core/CL/ICLArray.h"
 #include "arm_compute/core/CL/ICLMultiHOG.h"
-#include "arm_compute/core/CL/kernels/CLHOGDescriptorKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPDetectionWindowNonMaximaSuppressionKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLHOGDetector.h"
@@ -39,6 +38,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLHOGOrientationBinningKernel;
+class CLHOGBlockNormalizationKernel;
 /** Basic function to detect multiple objects (or the same object at different scales) on the same input image using HOG. This function calls the following kernels:
  *
  * -# @ref CLHOGGradient
@@ -62,6 +64,8 @@ public:
     CLHOGMultiDetection(const CLHOGMultiDetection &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     CLHOGMultiDetection &operator=(const CLHOGMultiDetection &) = delete;
+    /** Default destructor */
+    ~CLHOGMultiDetection();
     /** Initialise the function's source, destination, detection window strides, border mode, threshold and non-maxima suppression
      *
      * @param[in, out] input                    Input tensor. Data type supported: U8
@@ -110,21 +114,21 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                                  _memory_group;
-    CLHOGGradient                                _gradient_kernel;
-    std::vector<CLHOGOrientationBinningKernel>   _orient_bin_kernel;
-    std::vector<CLHOGBlockNormalizationKernel>   _block_norm_kernel;
-    std::vector<CLHOGDetector>                   _hog_detect_kernel;
-    CPPDetectionWindowNonMaximaSuppressionKernel _non_maxima_kernel;
-    std::vector<CLTensor>                        _hog_space;
-    std::vector<CLTensor>                        _hog_norm_space;
-    ICLDetectionWindowArray                     *_detection_windows;
-    CLTensor                                     _mag;
-    CLTensor                                     _phase;
-    bool                                         _non_maxima_suppression;
-    size_t                                       _num_orient_bin_kernel;
-    size_t                                       _num_block_norm_kernel;
-    size_t                                       _num_hog_detect_kernel;
+    MemoryGroup                                                 _memory_group;
+    CLHOGGradient                                               _gradient_kernel;
+    std::vector<std::unique_ptr<CLHOGOrientationBinningKernel>> _orient_bin_kernel;
+    std::vector<std::unique_ptr<CLHOGBlockNormalizationKernel>> _block_norm_kernel;
+    std::vector<CLHOGDetector>                                  _hog_detect_kernel;
+    CPPDetectionWindowNonMaximaSuppressionKernel                _non_maxima_kernel;
+    std::vector<CLTensor>                                       _hog_space;
+    std::vector<CLTensor>                                       _hog_norm_space;
+    ICLDetectionWindowArray                                    *_detection_windows;
+    CLTensor                                                    _mag;
+    CLTensor                                                    _phase;
+    bool                                                        _non_maxima_suppression;
+    size_t                                                      _num_orient_bin_kernel;
+    size_t                                                      _num_block_norm_kernel;
+    size_t                                                      _num_hog_detect_kernel;
 };
 }
 
diff --git a/arm_compute/runtime/CL/functions/CLHarrisCorners.h b/arm_compute/runtime/CL/functions/CLHarrisCorners.h
index 326a895d39..c9c67f5a28 100644
--- a/arm_compute/runtime/CL/functions/CLHarrisCorners.h
+++ b/arm_compute/runtime/CL/functions/CLHarrisCorners.h
@@ -24,16 +24,13 @@
 #ifndef ARM_COMPUTE_CLHARRISCORNERS_H
 #define ARM_COMPUTE_CLHARRISCORNERS_H
 
-#include "arm_compute/runtime/IFunction.h"
-
 #include "arm_compute/core/CL/ICLArray.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLHarrisCornersKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPCornerCandidatesKernel.h"
 #include "arm_compute/core/CPP/kernels/CPPSortEuclideanDistanceKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLNonMaximaSuppression3x3.h"
+#include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
 #include <cstdint>
@@ -41,6 +38,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLHarrisScoreKernel;
+class CLFillBorderKernel;
 class ICLTensor;
 using ICLImage = ICLTensor;
 
@@ -66,6 +66,8 @@ public:
     CLHarrisCorners(const CLHarrisCorners &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     const CLHarrisCorners &operator=(const CLHarrisCorners &) = delete;
+    /** Default destructor */
+    ~CLHarrisCorners();
     /** Initialize the function's source, destination, conv and border_mode.
      *
      * @param[in,out] input                 Source image. Data types supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -104,21 +106,21 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                    _memory_group;          /**< Function's memory group */
-    std::unique_ptr<IFunction>     _sobel;                 /**< Sobel function */
-    CLHarrisScoreKernel            _harris_score;          /**< Harris score kernel */
-    CLNonMaximaSuppression3x3      _non_max_suppr;         /**< Non-maxima suppression function */
-    CPPCornerCandidatesKernel      _candidates;            /**< Sort kernel */
-    CPPSortEuclideanDistanceKernel _sort_euclidean;        /**< Euclidean distance kernel */
-    CLFillBorderKernel             _border_gx;             /**< Border handler before running harris score */
-    CLFillBorderKernel             _border_gy;             /**< Border handler before running harris score */
-    CLImage                        _gx;                    /**< Source image - Gx component */
-    CLImage                        _gy;                    /**< Source image - Gy component */
-    CLImage                        _score;                 /**< Source image - Harris score */
-    CLImage                        _nonmax;                /**< Source image - Non-Maxima suppressed image */
-    std::vector<InternalKeypoint>  _corners_list;          /**< Array of InternalKeypoint. It stores the potential corner candidates */
-    int32_t                        _num_corner_candidates; /**< Number of potential corner candidates */
-    ICLKeyPointArray              *_corners;               /**< Output corners array */
+    MemoryGroup                          _memory_group;          /**< Function's memory group */
+    std::unique_ptr<IFunction>           _sobel;                 /**< Sobel function */
+    std::unique_ptr<CLHarrisScoreKernel> _harris_score;          /**< Harris score kernel */
+    CLNonMaximaSuppression3x3            _non_max_suppr;         /**< Non-maxima suppression function */
+    CPPCornerCandidatesKernel            _candidates;            /**< Sort kernel */
+    CPPSortEuclideanDistanceKernel       _sort_euclidean;        /**< Euclidean distance kernel */
+    std::unique_ptr<CLFillBorderKernel>  _border_gx;             /**< Border handler before running harris score */
+    std::unique_ptr<CLFillBorderKernel>  _border_gy;             /**< Border handler before running harris score */
+    CLImage                              _gx;                    /**< Source image - Gx component */
+    CLImage                              _gy;                    /**< Source image - Gy component */
+    CLImage                              _score;                 /**< Source image - Harris score */
+    CLImage                              _nonmax;                /**< Source image - Non-Maxima suppressed image */
+    std::vector<InternalKeypoint>        _corners_list;          /**< Array of InternalKeypoint. It stores the potential corner candidates */
+    int32_t                              _num_corner_candidates; /**< Number of potential corner candidates */
+    ICLKeyPointArray                    *_corners;               /**< Output corners array */
 };
 }
 #endif /*ARM_COMPUTE_CLHARRISCORNERS_H */
diff --git a/arm_compute/runtime/CL/functions/CLHistogram.h b/arm_compute/runtime/CL/functions/CLHistogram.h
index 7fdb8a9022..164bd0a28a 100644
--- a/arm_compute/runtime/CL/functions/CLHistogram.h
+++ b/arm_compute/runtime/CL/functions/CLHistogram.h
@@ -24,8 +24,8 @@
 #ifndef ARM_COMPUTE_CLHISTOGRAM_H
 #define ARM_COMPUTE_CLHISTOGRAM_H
 
-#include "arm_compute/core/CL/kernels/CLHistogramKernel.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "src/core/CL/kernels/CLHistogramKernel.h"
 
 namespace arm_compute
 {
diff --git a/arm_compute/runtime/CL/functions/CLInstanceNormalizationLayer.h b/arm_compute/runtime/CL/functions/CLInstanceNormalizationLayer.h
index d7aa11cbc8..d41f3fedf6 100644
--- a/arm_compute/runtime/CL/functions/CLInstanceNormalizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLInstanceNormalizationLayer.h
@@ -24,11 +24,14 @@
 #ifndef ARM_COMPUTE_CLINSTANCENORMALIZATIONLAYER_H
 #define ARM_COMPUTE_CLINSTANCENORMALIZATIONLAYER_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to perform a Instance normalization.
  *
diff --git a/arm_compute/runtime/CL/functions/CLIntegralImage.h b/arm_compute/runtime/CL/functions/CLIntegralImage.h
index 6b10ede650..0ecdbde8fe 100644
--- a/arm_compute/runtime/CL/functions/CLIntegralImage.h
+++ b/arm_compute/runtime/CL/functions/CLIntegralImage.h
@@ -24,11 +24,15 @@
 #ifndef ARM_COMPUTE_CLINTEGRALIMAGE_H
 #define ARM_COMPUTE_CLINTEGRALIMAGE_H
 
-#include "arm_compute/core/CL/kernels/CLIntegralImageKernel.h"
 #include "arm_compute/runtime/IFunction.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLIntegralImageHorKernel;
+class CLIntegralImageVertKernel;
 class ICLTensor;
 
 /** Basic function to execute integral image. This function calls the following OpenCL kernels:
@@ -42,6 +46,12 @@ class CLIntegralImage : public IFunction
 public:
     /** Default Constructor. */
     CLIntegralImage();
+    /** Prevent instances of this class from being copied */
+    CLIntegralImage(const CLIntegralImage &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLIntegralImage &operator=(const CLIntegralImage &) = delete;
+    /** Default destructor */
+    ~CLIntegralImage();
     /** Initialise the function's source, destinations and border mode.
      *
      * @param[in]  input  Source tensor. Data types supported: U8.
@@ -60,8 +70,8 @@ public:
     void run() override;
 
 protected:
-    CLIntegralImageHorKernel  _integral_hor;  /**< Integral Image Horizontal kernel */
-    CLIntegralImageVertKernel _integral_vert; /**< Integral Image Vertical kernel */
+    std::unique_ptr<CLIntegralImageHorKernel>  _integral_hor;  /**< Integral Image Horizontal kernel */
+    std::unique_ptr<CLIntegralImageVertKernel> _integral_vert; /**< Integral Image Vertical kernel */
 };
 }
 #endif /*ARM_COMPUTE_CLINTEGRALIMAGE_H */
diff --git a/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h b/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h
index bc79101d9d..401d249eb4 100644
--- a/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h
+++ b/arm_compute/runtime/CL/functions/CLL2NormalizeLayer.h
@@ -24,7 +24,6 @@
 #ifndef ARM_COMPUTE_CLL2NORMALIZELAYER_H
 #define ARM_COMPUTE_CLL2NORMALIZELAYER_H
 
-#include "arm_compute/core/CL/kernels/CLL2NormalizeLayerKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
@@ -37,7 +36,10 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLL2NormalizeLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to perform a L2 normalization on a given axis.
  *
@@ -50,6 +52,16 @@ class CLL2NormalizeLayer : public IFunction
 public:
     /** Constructor */
     CLL2NormalizeLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Default Destructor */
+    ~CLL2NormalizeLayer();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLL2NormalizeLayer(const CLL2NormalizeLayer &) = delete;
+    /** Default move constructor */
+    CLL2NormalizeLayer(CLL2NormalizeLayer &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLL2NormalizeLayer &operator=(const CLL2NormalizeLayer &) = delete;
+    /** Default move assignment operator */
+    CLL2NormalizeLayer &operator=(CLL2NormalizeLayer &&) = default;
 
     /** Set the input and output tensors.
      *
@@ -84,10 +96,10 @@ public:
     void run() override;
 
 private:
-    MemoryGroup              _memory_group;
-    CLReductionOperation     _reduce_func;
-    CLL2NormalizeLayerKernel _normalize_kernel;
-    CLTensor                 _sumsq;
+    MemoryGroup                               _memory_group;
+    CLReductionOperation                      _reduce_func;
+    std::unique_ptr<CLL2NormalizeLayerKernel> _normalize_kernel;
+    CLTensor                                  _sumsq;
 };
 }
 #endif /*ARM_COMPUTE_CLL2NORMALIZELAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLLSTMLayer.h b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
index 1a8b33463d..017f26aa1e 100644
--- a/arm_compute/runtime/CL/functions/CLLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
@@ -26,8 +26,6 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLCopyKernel.h"
-#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
@@ -45,6 +43,10 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLCopyKernel;
+class CLMemsetKernel;
+class CLTransposeKernel;
 class ICLTensor;
 
 /** This function performs a single time step in a Long Short-Term Memory (LSTM) layer.
@@ -55,6 +57,16 @@ class CLLSTMLayer : public IFunction
 public:
     /** Default constructor */
     CLLSTMLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLLSTMLayer(const CLLSTMLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLLSTMLayer &operator=(const CLLSTMLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLLSTMLayer(CLLSTMLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLLSTMLayer &operator=(CLLSTMLayer &&) = delete;
+    /** Default destructor */
+    ~CLLSTMLayer();
     /** Initialize function's tensors.
      *
      * @param[in]  input                       Source tensor. Input is a 2D tensor with dimensions [input_size, batch_size]. Data types supported: F16/F32.
@@ -200,90 +212,90 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup                    _memory_group;
-    CLFullyConnectedLayer          _fully_connected_input_gate;
-    CLArithmeticAddition           _accum_input_gate1;
-    CLArithmeticSubtraction        _subtract_input_gate;
-    CLPixelWiseMultiplication      _pixelwise_mul_input_gate;
-    CLActivationLayer              _activation_input_gate;
-    CLFullyConnectedLayer          _fully_connected_forget_gate;
-    CLArithmeticAddition           _accum_forget_gate1;
-    CLPixelWiseMultiplication      _pixelwise_mul_forget_gate;
-    CLActivationLayer              _activation_forget_gate;
-    CLFullyConnectedLayer          _fully_connected_cell_state;
-    CLGEMM                         _gemm_cell_state1;
-    CLTransposeKernel              _transpose_cell_state;
-    CLArithmeticAddition           _accum_cell_state1;
-    CLArithmeticAddition           _accum_cell_state2;
-    CLPixelWiseMultiplication      _pixelwise_mul_cell_state1;
-    CLActivationLayer              _activation_cell_state;
-    CLActivationLayer              _cell_clip;
-    CLPixelWiseMultiplication      _pixelwise_mul_cell_state2;
-    CLFullyConnectedLayer          _fully_connected_output;
-    CLPixelWiseMultiplication      _pixelwise_mul_output_state1;
-    CLArithmeticAddition           _accum_output1;
-    CLActivationLayer              _activation_output;
-    CLActivationLayer              _activation_output_state;
-    CLPixelWiseMultiplication      _pixelwise_mul_output_state2;
-    CLFullyConnectedLayer          _fully_connected_output_state;
-    CLActivationLayer              _projection_clip;
-    CLCopyKernel                   _copy_cell_state;
-    CLCopyKernel                   _copy_output;
-    CLConcatenateLayer             _concat_scratch_buffer;
-    CLConcatenateLayer             _concat_inputs_forget_gate;
-    CLConcatenateLayer             _concat_weights_forget_gate;
-    CLConcatenateLayer             _concat_weights_input_gate;
-    CLConcatenateLayer             _concat_weights_output;
-    CLMemsetKernel                 _ones_memset_kernel;
-    CLMeanStdDevNormalizationLayer _mean_std_norm_input_gate;
-    CLPixelWiseMultiplication      _pixelwise_mul_input_gate_coeff;
-    CLArithmeticAddition           _accum_input_gate_bias;
-    CLMeanStdDevNormalizationLayer _mean_std_norm_forget_gate;
-    CLPixelWiseMultiplication      _pixelwise_mul_forget_gate_coeff;
-    CLArithmeticAddition           _accum_forget_gate_bias;
-    CLMeanStdDevNormalizationLayer _mean_std_norm_cell_gate;
-    CLPixelWiseMultiplication      _pixelwise_mul_cell_gate_coeff;
-    CLArithmeticAddition           _accum_cell_gate_bias;
-    CLMeanStdDevNormalizationLayer _mean_std_norm_output_gate;
-    CLPixelWiseMultiplication      _pixelwise_mul_output_gate_coeff;
-    CLArithmeticAddition           _accum_output_gate_bias;
-    CLTensor                       _input_gate_out1;
-    CLTensor                       _input_gate_out2;
-    CLTensor                       _input_gate_out3;
-    CLTensor                       _input_gate_out4;
-    CLTensor                       _forget_gate_out1;
-    CLTensor                       _forget_gate_out2;
-    CLTensor                       _forget_gate_out3;
-    CLTensor                       _forget_gate_out4;
-    CLTensor                       _forget_gate_out5;
-    CLTensor                       _forget_gate_out6;
-    CLTensor                       _cell_state_out1;
-    CLTensor                       _cell_state_out2;
-    CLTensor                       _cell_state_out3;
-    CLTensor                       _cell_state_out4;
-    CLTensor                       _cell_state_out5;
-    CLTensor                       _output1;
-    CLTensor                       _output2;
-    CLTensor                       _output3;
-    CLTensor                       _output4;
-    CLTensor                       _cell_state_activation;
-    CLTensor                       _output_state1;
-    CLTensor                       _ones;
-    CLTensor                       _input_layer_norm_out1;
-    CLTensor                       _input_layer_norm_out2;
-    CLTensor                       _forget_layer_norm_out1;
-    CLTensor                       _forget_layer_norm_out2;
-    CLTensor                       _cell_layer_norm_out1;
-    CLTensor                       _cell_layer_norm_out2;
-    CLTensor                       _output_layer_norm_out1;
-    CLTensor                       _output_layer_norm_out2;
-    bool                           _run_peephole_opt;
-    bool                           _run_cifg_opt;
-    bool                           _perform_cell_clipping;
-    bool                           _has_projection_weights;
-    bool                           _perform_projection_clipping;
-    bool                           _is_prepared;
-    bool                           _is_layer_norm_lstm;
+    MemoryGroup                        _memory_group;
+    CLFullyConnectedLayer              _fully_connected_input_gate;
+    CLArithmeticAddition               _accum_input_gate1;
+    CLArithmeticSubtraction            _subtract_input_gate;
+    CLPixelWiseMultiplication          _pixelwise_mul_input_gate;
+    CLActivationLayer                  _activation_input_gate;
+    CLFullyConnectedLayer              _fully_connected_forget_gate;
+    CLArithmeticAddition               _accum_forget_gate1;
+    CLPixelWiseMultiplication          _pixelwise_mul_forget_gate;
+    CLActivationLayer                  _activation_forget_gate;
+    CLFullyConnectedLayer              _fully_connected_cell_state;
+    CLGEMM                             _gemm_cell_state1;
+    std::unique_ptr<CLTransposeKernel> _transpose_cell_state;
+    CLArithmeticAddition               _accum_cell_state1;
+    CLArithmeticAddition               _accum_cell_state2;
+    CLPixelWiseMultiplication          _pixelwise_mul_cell_state1;
+    CLActivationLayer                  _activation_cell_state;
+    CLActivationLayer                  _cell_clip;
+    CLPixelWiseMultiplication          _pixelwise_mul_cell_state2;
+    CLFullyConnectedLayer              _fully_connected_output;
+    CLPixelWiseMultiplication          _pixelwise_mul_output_state1;
+    CLArithmeticAddition               _accum_output1;
+    CLActivationLayer                  _activation_output;
+    CLActivationLayer                  _activation_output_state;
+    CLPixelWiseMultiplication          _pixelwise_mul_output_state2;
+    CLFullyConnectedLayer              _fully_connected_output_state;
+    CLActivationLayer                  _projection_clip;
+    std::unique_ptr<CLCopyKernel>      _copy_cell_state;
+    std::unique_ptr<CLCopyKernel>      _copy_output;
+    CLConcatenateLayer                 _concat_scratch_buffer;
+    CLConcatenateLayer                 _concat_inputs_forget_gate;
+    CLConcatenateLayer                 _concat_weights_forget_gate;
+    CLConcatenateLayer                 _concat_weights_input_gate;
+    CLConcatenateLayer                 _concat_weights_output;
+    std::unique_ptr<CLMemsetKernel>    _ones_memset_kernel;
+    CLMeanStdDevNormalizationLayer     _mean_std_norm_input_gate;
+    CLPixelWiseMultiplication          _pixelwise_mul_input_gate_coeff;
+    CLArithmeticAddition               _accum_input_gate_bias;
+    CLMeanStdDevNormalizationLayer     _mean_std_norm_forget_gate;
+    CLPixelWiseMultiplication          _pixelwise_mul_forget_gate_coeff;
+    CLArithmeticAddition               _accum_forget_gate_bias;
+    CLMeanStdDevNormalizationLayer     _mean_std_norm_cell_gate;
+    CLPixelWiseMultiplication          _pixelwise_mul_cell_gate_coeff;
+    CLArithmeticAddition               _accum_cell_gate_bias;
+    CLMeanStdDevNormalizationLayer     _mean_std_norm_output_gate;
+    CLPixelWiseMultiplication          _pixelwise_mul_output_gate_coeff;
+    CLArithmeticAddition               _accum_output_gate_bias;
+    CLTensor                           _input_gate_out1;
+    CLTensor                           _input_gate_out2;
+    CLTensor                           _input_gate_out3;
+    CLTensor                           _input_gate_out4;
+    CLTensor                           _forget_gate_out1;
+    CLTensor                           _forget_gate_out2;
+    CLTensor                           _forget_gate_out3;
+    CLTensor                           _forget_gate_out4;
+    CLTensor                           _forget_gate_out5;
+    CLTensor                           _forget_gate_out6;
+    CLTensor                           _cell_state_out1;
+    CLTensor                           _cell_state_out2;
+    CLTensor                           _cell_state_out3;
+    CLTensor                           _cell_state_out4;
+    CLTensor                           _cell_state_out5;
+    CLTensor                           _output1;
+    CLTensor                           _output2;
+    CLTensor                           _output3;
+    CLTensor                           _output4;
+    CLTensor                           _cell_state_activation;
+    CLTensor                           _output_state1;
+    CLTensor                           _ones;
+    CLTensor                           _input_layer_norm_out1;
+    CLTensor                           _input_layer_norm_out2;
+    CLTensor                           _forget_layer_norm_out1;
+    CLTensor                           _forget_layer_norm_out2;
+    CLTensor                           _cell_layer_norm_out1;
+    CLTensor                           _cell_layer_norm_out2;
+    CLTensor                           _output_layer_norm_out1;
+    CLTensor                           _output_layer_norm_out2;
+    bool                               _run_peephole_opt;
+    bool                               _run_cifg_opt;
+    bool                               _perform_cell_clipping;
+    bool                               _has_projection_weights;
+    bool                               _perform_projection_clipping;
+    bool                               _is_prepared;
+    bool                               _is_layer_norm_lstm;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLLSTMLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
index ba85c6140c..3bbf9f2c30 100644
--- a/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLocallyConnectedLayer.h
@@ -26,10 +26,6 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLCol2ImKernel.h"
-#include "arm_compute/core/CL/kernels/CLIm2ColKernel.h"
-#include "arm_compute/core/CL/kernels/CLLocallyConnectedMatrixMultiplyKernel.h"
-#include "arm_compute/core/CL/kernels/CLWeightsReshapeKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IMemoryManager.h"
@@ -39,7 +35,13 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLCol2ImKernel;
+class CLIm2ColKernel;
+class CLWeightsReshapeKernel;
+class CLLocallyConnectedMatrixMultiplyKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to compute the locally connected layer. This function calls the following OpenCL kernels:
  *
@@ -108,16 +110,16 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup                            _memory_group;
-    CLIm2ColKernel                         _input_im2col_kernel;
-    CLWeightsReshapeKernel                 _weights_reshape_kernel;
-    CLLocallyConnectedMatrixMultiplyKernel _mm_kernel;
-    CLCol2ImKernel                         _output_col2im_kernel;
-    CLTensor                               _input_im2col_reshaped;
-    CLTensor                               _weights_reshaped;
-    CLTensor                               _gemm_output;
-    bool                                   _is_prepared;
-    const ICLTensor                       *_original_weights;
+    MemoryGroup                                             _memory_group;
+    std::unique_ptr<CLIm2ColKernel>                         _input_im2col_kernel;
+    std::unique_ptr<CLWeightsReshapeKernel>                 _weights_reshape_kernel;
+    std::unique_ptr<CLLocallyConnectedMatrixMultiplyKernel> _mm_kernel;
+    std::unique_ptr<CLCol2ImKernel>                         _output_col2im_kernel;
+    CLTensor                                                _input_im2col_reshaped;
+    CLTensor                                                _weights_reshaped;
+    CLTensor                                                _gemm_output;
+    bool                                                    _is_prepared;
+    const ICLTensor                                        *_original_weights;
 };
 }
 #endif /* ARM_COMPUTE_CLLOCALLYCONNECTEDLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLMagnitude.h b/arm_compute/runtime/CL/functions/CLMagnitude.h
index ad7cc778e5..6ac141641c 100644
--- a/arm_compute/runtime/CL/functions/CLMagnitude.h
+++ b/arm_compute/runtime/CL/functions/CLMagnitude.h
@@ -29,6 +29,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLMagnitudePhaseKernel. */
diff --git a/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h b/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
index 5c8548f9e0..693862fb89 100644
--- a/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
+++ b/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
@@ -24,14 +24,19 @@
 #ifndef ARM_COMPUTE_CLMAXUNPOOLINGLAYER_H
 #define ARM_COMPUTE_CLMAXUNPOOLINGLAYER_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLMaxUnpoolingLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
+#include <memory>
 
 namespace arm_compute
 {
-class ITensor;
+class CLCompileContext;
+class ICLTensor;
+class ITensorInfo;
+class CLMaxUnpoolingLayerKernel;
+class CLMemsetKernel;
+struct PoolingLayerInfo;
 
 /** Function to perform MaxUnpooling. This function calls the following OpenCL kernels:
  *
@@ -43,6 +48,12 @@ class CLMaxUnpoolingLayer : public IFunction
 public:
     /** Constructor */
     CLMaxUnpoolingLayer();
+    /** Prevent instances of this class from being copied */
+    CLMaxUnpoolingLayer(const CLMaxUnpoolingLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLMaxUnpoolingLayer &operator=(const CLMaxUnpoolingLayer &) = delete;
+    /** Default destructor */
+    ~CLMaxUnpoolingLayer();
     /** Set the input and output tensors.
      *
      * @note Output shape must be equal to the shape of the original input to pool.
@@ -88,8 +99,8 @@ public:
     void run() override;
 
 private:
-    CLMemsetKernel            _memset_kernel;
-    CLMaxUnpoolingLayerKernel _unpooling_layer_kernel;
+    std::unique_ptr<CLMemsetKernel>            _memset_kernel;
+    std::unique_ptr<CLMaxUnpoolingLayerKernel> _unpooling_layer_kernel;
 };
 }
 #endif /* ARM_COMPUTE_CLMAXUNPOOLINGLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLMeanStdDev.h b/arm_compute/runtime/CL/functions/CLMeanStdDev.h
index be192a7c11..d9ced1393e 100644
--- a/arm_compute/runtime/CL/functions/CLMeanStdDev.h
+++ b/arm_compute/runtime/CL/functions/CLMeanStdDev.h
@@ -25,15 +25,20 @@
 #define ARM_COMPUTE_CLMEANSTDDEV_H
 
 #include "arm_compute/core/CL/OpenCL.h"
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLMeanStdDevKernel.h"
 #include "arm_compute/runtime/CL/functions/CLReductionOperation.h"
 #include "arm_compute/runtime/IFunction.h"
 #include "arm_compute/runtime/IMemoryManager.h"
 #include "arm_compute/runtime/MemoryGroup.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class ICLTensor;
+class ITensorInfo;
+class CLFillBorderKernel;
+class CLMeanStdDevKernel;
 /** Basic function to execute mean and standard deviation by calling @ref CLMeanStdDevKernel */
 class CLMeanStdDev : public IFunction
 {
@@ -49,7 +54,7 @@ public:
     /** Allow instances of this class to be moved */
     CLMeanStdDev &operator=(CLMeanStdDev &&) = default;
     /** Default destructor */
-    ~CLMeanStdDev() = default;
+    ~CLMeanStdDev();
     /** Initialise the kernel's inputs and outputs.
      *
      * @param[in, out] input  Input image. Data types supported: U8/F16/F32. (Written to only for border filling)
@@ -83,20 +88,20 @@ private:
     void run_float();
     void run_int();
 
-    MemoryGroup          _memory_group;               /**< Function's memory group */
-    DataType             _data_type;                  /**< Input data type. */
-    unsigned int         _num_pixels;                 /**< Number of image's pixels. */
-    bool                 _run_stddev;                 /**< Flag for knowing if we should run stddev reduction function. */
-    CLReductionOperation _reduction_operation_mean;   /**< Reduction operation function for computing mean value. */
-    CLReductionOperation _reduction_operation_stddev; /**< Reduction operation function for computing standard deviation. */
-    CLTensor             _reduction_output_mean;      /**< Reduction operation output tensor for mean value. */
-    CLTensor             _reduction_output_stddev;    /**< Reduction operation output tensor for standard deviation value. */
-    float               *_mean;                       /**< Pointer that holds the mean value. */
-    float               *_stddev;                     /**< Pointer that holds the standard deviation value. */
-    CLMeanStdDevKernel   _mean_stddev_kernel;         /**< Kernel that standard deviation calculation. */
-    CLFillBorderKernel   _fill_border_kernel;         /**< Kernel that fills the border with zeroes. */
-    cl::Buffer           _global_sum;                 /**< Variable that holds the global sum among calls in order to ease reduction */
-    cl::Buffer           _global_sum_squared;         /**< Variable that holds the global sum of squared values among calls in order to ease reduction */
+    MemoryGroup                         _memory_group;               /**< Function's memory group */
+    DataType                            _data_type;                  /**< Input data type. */
+    unsigned int                        _num_pixels;                 /**< Number of image's pixels. */
+    bool                                _run_stddev;                 /**< Flag for knowing if we should run stddev reduction function. */
+    CLReductionOperation                _reduction_operation_mean;   /**< Reduction operation function for computing mean value. */
+    CLReductionOperation                _reduction_operation_stddev; /**< Reduction operation function for computing standard deviation. */
+    CLTensor                            _reduction_output_mean;      /**< Reduction operation output tensor for mean value. */
+    CLTensor                            _reduction_output_stddev;    /**< Reduction operation output tensor for standard deviation value. */
+    float                              *_mean;                       /**< Pointer that holds the mean value. */
+    float                              *_stddev;                     /**< Pointer that holds the standard deviation value. */
+    std::unique_ptr<CLMeanStdDevKernel> _mean_stddev_kernel;         /**< Kernel that standard deviation calculation. */
+    std::unique_ptr<CLFillBorderKernel> _fill_border_kernel;         /**< Kernel that fills the border with zeroes. */
+    cl::Buffer                          _global_sum;                 /**< Variable that holds the global sum among calls in order to ease reduction */
+    cl::Buffer                          _global_sum_squared;         /**< Variable that holds the global sum of squared values among calls in order to ease reduction */
 };
 }
 #endif /*ARM_COMPUTE_CLMEANSTDDEV_H */
diff --git a/arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h b/arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h
index 1627de1ae8..cfe59eac09 100644
--- a/arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute mean and standard deviation normalization by calling @ref CLMeanStdDevNormalizationKernel */
 class CLMeanStdDevNormalizationLayer : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLMedian3x3.h b/arm_compute/runtime/CL/functions/CLMedian3x3.h
index 7f67f958c1..6c0458203e 100644
--- a/arm_compute/runtime/CL/functions/CLMedian3x3.h
+++ b/arm_compute/runtime/CL/functions/CLMedian3x3.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute median filter. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLMinMaxLocation.h b/arm_compute/runtime/CL/functions/CLMinMaxLocation.h
index 04926f7bd0..4e3f28b006 100644
--- a/arm_compute/runtime/CL/functions/CLMinMaxLocation.h
+++ b/arm_compute/runtime/CL/functions/CLMinMaxLocation.h
@@ -24,12 +24,16 @@
 #ifndef ARM_COMPUTE_CLMINMAXLOCATION_H
 #define ARM_COMPUTE_CLMINMAXLOCATION_H
 
-#include "arm_compute/core/CL/kernels/CLMinMaxLocationKernel.h"
 #include "arm_compute/runtime/CL/CLArray.h"
 #include "arm_compute/runtime/IFunction.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLMinMaxKernel;
+class CLMinMaxLocationKernel;
 class ICLTensor;
 using ICLImage = ICLTensor;
 
@@ -51,6 +55,8 @@ public:
     CLMinMaxLocation(CLMinMaxLocation &&) = default;
     /** Allow instances of this class to be moved */
     CLMinMaxLocation &operator=(CLMinMaxLocation &&) = default;
+    /** Default destructor */
+    ~CLMinMaxLocation();
     /** Initialise the kernel's inputs and outputs.
      *
      * @note When locations of min and max occurrences are requested, the reported number of locations is limited to the given array size.
@@ -87,16 +93,16 @@ public:
     void run() override;
 
 private:
-    CLMinMaxKernel         _min_max_kernel;     /**< Kernel that performs min/max */
-    CLMinMaxLocationKernel _min_max_loc_kernel; /**< Kernel that counts min/max occurrences and identifies their positions */
-    cl::Buffer             _min_max_vals;       /**< Buffer to collect min, max values */
-    cl::Buffer             _min_max_count_vals; /**< Buffer to collect min, max values */
-    void                  *_min;                /**< Minimum value. */
-    void                  *_max;                /**< Maximum value. */
-    uint32_t              *_min_count;          /**< Minimum value occurrences. */
-    uint32_t              *_max_count;          /**< Maximum value occurrences. */
-    CLCoordinates2DArray *_min_loc;             /**< Minimum value occurrences coordinates. */
-    CLCoordinates2DArray *_max_loc;             /**< Maximum value occurrences  coordinates. */
+    std::unique_ptr<CLMinMaxKernel>         _min_max_kernel;     /**< Kernel that performs min/max */
+    std::unique_ptr<CLMinMaxLocationKernel> _min_max_loc_kernel; /**< Kernel that counts min/max occurrences and identifies their positions */
+    cl::Buffer                              _min_max_vals;       /**< Buffer to collect min, max values */
+    cl::Buffer                              _min_max_count_vals; /**< Buffer to collect min, max values */
+    void                                   *_min;                /**< Minimum value. */
+    void                                   *_max;                /**< Maximum value. */
+    uint32_t                               *_min_count;          /**< Minimum value occurrences. */
+    uint32_t                               *_max_count;          /**< Maximum value occurrences. */
+    CLCoordinates2DArray                   *_min_loc;            /**< Minimum value occurrences coordinates. */
+    CLCoordinates2DArray                   *_max_loc;            /**< Maximum value occurrences  coordinates. */
 };
 }
 #endif /*ARM_COMPUTE_CLMINMAXLOCATION_H */
diff --git a/arm_compute/runtime/CL/functions/CLNonLinearFilter.h b/arm_compute/runtime/CL/functions/CLNonLinearFilter.h
index 8b7e350e09..1b466bf662 100644
--- a/arm_compute/runtime/CL/functions/CLNonLinearFilter.h
+++ b/arm_compute/runtime/CL/functions/CLNonLinearFilter.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute non linear filter. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLNonMaximaSuppression3x3.h b/arm_compute/runtime/CL/functions/CLNonMaximaSuppression3x3.h
index 556de1c64c..c767a042ff 100644
--- a/arm_compute/runtime/CL/functions/CLNonMaximaSuppression3x3.h
+++ b/arm_compute/runtime/CL/functions/CLNonMaximaSuppression3x3.h
@@ -29,6 +29,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute non-maxima suppression over a 3x3 window. This function calls the following CL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLNormalizationLayer.h b/arm_compute/runtime/CL/functions/CLNormalizationLayer.h
index a2d46b368f..389b21e5c8 100644
--- a/arm_compute/runtime/CL/functions/CLNormalizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLNormalizationLayer.h
@@ -24,18 +24,19 @@
 #ifndef ARM_COMPUTE_CLNORMALIZATIONLAYER_H
 #define ARM_COMPUTE_CLNORMALIZATIONLAYER_H
 
-#include "arm_compute/runtime/IFunction.h"
-
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLNormalizationLayerKernel.h"
-#include "arm_compute/core/CL/kernels/CLPixelWiseMultiplicationKernel.h"
+#include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/Types.h"
+#include <memory>
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLNormalizationLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to compute a normalization layer. This function calls the following CL kernels:
  *
@@ -48,6 +49,16 @@ class CLNormalizationLayer : public IFunction
 public:
     /** Default constructor */
     CLNormalizationLayer();
+    /** Prevent instances of this class from being copied */
+    CLNormalizationLayer(const CLNormalizationLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLNormalizationLayer &operator=(const CLNormalizationLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLNormalizationLayer(CLNormalizationLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLNormalizationLayer &operator=(CLNormalizationLayer &&) = delete;
+    /** Default destructor */
+    ~CLNormalizationLayer();
     /** Set the input and output tensors.
      *
      * @param[in, out] input     Source tensor. 3 lower dims represent a single input with dimensions [width, height, IFM],
@@ -85,8 +96,8 @@ public:
     void run() override;
 
 private:
-    CLNormalizationLayerKernel _norm_kernel;    /**< Normalization layer kernel to run */
-    CLFillBorderKernel         _border_handler; /**< Kernel to handle  borders */
+    std::unique_ptr<CLNormalizationLayerKernel> _norm_kernel;    /**< Normalization layer kernel to run */
+    std::unique_ptr<CLFillBorderKernel>         _border_handler; /**< Kernel to handle  borders */
 };
 }
 #endif /* ARM_COMPUTE_CLNORMALIZATIONLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLNormalizePlanarYUVLayer.h b/arm_compute/runtime/CL/functions/CLNormalizePlanarYUVLayer.h
index cf4a9b6497..de5155c65a 100644
--- a/arm_compute/runtime/CL/functions/CLNormalizePlanarYUVLayer.h
+++ b/arm_compute/runtime/CL/functions/CLNormalizePlanarYUVLayer.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLNormalizePlanarYUVLayerKernel
  *
diff --git a/arm_compute/runtime/CL/functions/CLOpticalFlow.h b/arm_compute/runtime/CL/functions/CLOpticalFlow.h
index adce6748c8..0e34374aa5 100644
--- a/arm_compute/runtime/CL/functions/CLOpticalFlow.h
+++ b/arm_compute/runtime/CL/functions/CLOpticalFlow.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLOPTICALFLOW_H
 #define ARM_COMPUTE_CLOPTICALFLOW_H
 
-#include "arm_compute/core/CL/kernels/CLLKTrackerKernel.h"
-
 #include "arm_compute/core/IArray.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLArray.h"
@@ -41,7 +39,12 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class CLPyramid;
+class CLLKTrackerInitKernel;
+class CLLKTrackerStage0Kernel;
+class CLLKTrackerStage1Kernel;
+class CLLKTrackerFinalizeKernel;
 
 /** OpenCL Array of Internal Keypoints */
 using CLLKInternalKeypointArray = CLArray<CLLKInternalKeypoint>;
@@ -71,6 +74,8 @@ public:
     CLOpticalFlow(CLOpticalFlow &&) = default;
     /** Allow instances of this class to be moved */
     CLOpticalFlow &operator=(CLOpticalFlow &&) = default;
+    /** Default destructor */
+    ~CLOpticalFlow();
     /**  Initialise the function input and output
      *
      * @param[in]  old_pyramid           Pointer to the pyramid for the old tensor. Data types supported U8
@@ -117,22 +122,22 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                                _memory_group;
-    std::vector<CLLKTrackerInitKernel>         _tracker_init_kernel;
-    std::vector<CLLKTrackerStage0Kernel>       _tracker_stage0_kernel;
-    std::vector<CLLKTrackerStage1Kernel>       _tracker_stage1_kernel;
-    CLLKTrackerFinalizeKernel                  _tracker_finalize_kernel;
-    std::vector<CLScharr3x3>                   _func_scharr;
-    std::vector<CLTensor>                      _scharr_gx;
-    std::vector<CLTensor>                      _scharr_gy;
-    const ICLKeyPointArray                    *_old_points;
-    const ICLKeyPointArray                    *_new_points_estimates;
-    ICLKeyPointArray                          *_new_points;
-    std::unique_ptr<CLLKInternalKeypointArray> _old_points_internal;
-    std::unique_ptr<CLLKInternalKeypointArray> _new_points_internal;
-    std::unique_ptr<CLCoefficientTableArray>   _coefficient_table;
-    std::unique_ptr<CLOldValueArray>           _old_values;
-    size_t                                     _num_levels;
+    MemoryGroup                                           _memory_group;
+    std::vector<std::unique_ptr<CLLKTrackerInitKernel>>   _tracker_init_kernel;
+    std::vector<std::unique_ptr<CLLKTrackerStage0Kernel>> _tracker_stage0_kernel;
+    std::vector<std::unique_ptr<CLLKTrackerStage1Kernel>> _tracker_stage1_kernel;
+    std::unique_ptr<CLLKTrackerFinalizeKernel>            _tracker_finalize_kernel;
+    std::vector<CLScharr3x3>                              _func_scharr;
+    std::vector<CLTensor>                                 _scharr_gx;
+    std::vector<CLTensor>                                 _scharr_gy;
+    const ICLKeyPointArray                               *_old_points;
+    const ICLKeyPointArray                               *_new_points_estimates;
+    ICLKeyPointArray                                     *_new_points;
+    std::unique_ptr<CLLKInternalKeypointArray>            _old_points_internal;
+    std::unique_ptr<CLLKInternalKeypointArray>            _new_points_internal;
+    std::unique_ptr<CLCoefficientTableArray>              _coefficient_table;
+    std::unique_ptr<CLOldValueArray>                      _old_values;
+    size_t                                                _num_levels;
 };
 }
 #endif /*ARM_COMPUTE_CLOPTICALFLOW_H */
diff --git a/arm_compute/runtime/CL/functions/CLPReluLayer.h b/arm_compute/runtime/CL/functions/CLPReluLayer.h
index ffde9ec186..ab32bccc24 100644
--- a/arm_compute/runtime/CL/functions/CLPReluLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPReluLayer.h
@@ -24,13 +24,14 @@
 #ifndef ARM_COMPUTE_CLPRELULAYER_H
 #define ARM_COMPUTE_CLPRELULAYER_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/runtime/CL/ICLOperator.h"
 #include "arm_compute/runtime/IFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 namespace experimental
 {
diff --git a/arm_compute/runtime/CL/functions/CLPadLayer.h b/arm_compute/runtime/CL/functions/CLPadLayer.h
index e3a923f81c..2bbde30fc2 100644
--- a/arm_compute/runtime/CL/functions/CLPadLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPadLayer.h
@@ -24,13 +24,15 @@
 #ifndef ARM_COMPUTE_CLPADLAYER_H
 #define ARM_COMPUTE_CLPADLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLCopyKernel.h"
-#include "arm_compute/core/CL/kernels/CLPadLayerKernel.h"
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLPadLayerKernel;
+class CLCopyKernel;
 class ICLTensor;
 
 /** Basic function to pad a tensor. This function calls the following OpenCL functions/kernels:
@@ -51,6 +53,8 @@ public:
     CLPadLayer &operator=(const CLPadLayer &) = delete;
     /** Default move assignment operator */
     CLPadLayer &operator=(CLPadLayer &&) = default;
+    /** Default destructor */
+    ~CLPadLayer();
 
     /** Initialize the function
      *
@@ -95,9 +99,9 @@ public:
 private:
     void configure_reflect_mode(ICLTensor *input, ICLTensor *output);
 
-    CLPadLayerKernel _pad_kernel;
-    CLCopyKernel     _copy_kernel;
-    bool             _perform_pad;
+    std::unique_ptr<CLPadLayerKernel> _pad_kernel;
+    std::unique_ptr<CLCopyKernel>     _copy_kernel;
+    bool                              _perform_pad;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_PADLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLPermute.h b/arm_compute/runtime/CL/functions/CLPermute.h
index abc23eff0c..50e81da7c4 100644
--- a/arm_compute/runtime/CL/functions/CLPermute.h
+++ b/arm_compute/runtime/CL/functions/CLPermute.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute an @ref CLPermuteKernel. */
 class CLPermute : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLPhase.h b/arm_compute/runtime/CL/functions/CLPhase.h
index 2731a08a52..34b8e72175 100644
--- a/arm_compute/runtime/CL/functions/CLPhase.h
+++ b/arm_compute/runtime/CL/functions/CLPhase.h
@@ -29,6 +29,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute an @ref CLMagnitudePhaseKernel. */
diff --git a/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h b/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h
index 2066012306..6432cd040d 100644
--- a/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h
+++ b/arm_compute/runtime/CL/functions/CLPixelWiseMultiplication.h
@@ -24,14 +24,16 @@
 #ifndef ARM_COMPUTE_CLPIXELWISEMULTIPLICATION_H
 #define ARM_COMPUTE_CLPIXELWISEMULTIPLICATION_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
 #include "arm_compute/runtime/CL/ICLOperator.h"
 #include "arm_compute/runtime/IFunction.h"
 
 namespace arm_compute
 {
 // Forward declaration
+class CLCompileContext;
+class CLFillBorderKernel;
 class ICLTensor;
+class ITensorInfo;
 
 namespace experimental
 {
@@ -106,7 +108,7 @@ public:
     void run(ITensorPack &tensors) override;
 
 private:
-    CLFillBorderKernel _border_handler;
+    std::unique_ptr<CLFillBorderKernel> _border_handler;
 };
 
 /** Basic function to run @ref CLComplexPixelWiseMultiplicationKernel. */
@@ -139,7 +141,7 @@ public:
     void run(ITensorPack &tensors) override;
 
 private:
-    CLFillBorderKernel _border_handler;
+    std::unique_ptr<CLFillBorderKernel> _border_handler;
 };
 } // namespace experimental
 
diff --git a/arm_compute/runtime/CL/functions/CLPoolingLayer.h b/arm_compute/runtime/CL/functions/CLPoolingLayer.h
index 96dacf9322..ef1f426c22 100644
--- a/arm_compute/runtime/CL/functions/CLPoolingLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPoolingLayer.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to simulate a pooling layer with the specified pooling operation. This function calls the following OpenCL kernels:
  *
diff --git a/arm_compute/runtime/CL/functions/CLPriorBoxLayer.h b/arm_compute/runtime/CL/functions/CLPriorBoxLayer.h
index 9a78e77307..9129bfd064 100644
--- a/arm_compute/runtime/CL/functions/CLPriorBoxLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPriorBoxLayer.h
@@ -24,13 +24,16 @@
 #ifndef ARM_COMPUTE_CLPRIORBOXLAYER_H
 #define ARM_COMPUTE_CLPRIORBOXLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLPriorBoxLayerKernel.h"
+#include "arm_compute/core/CL/OpenCL.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLPriorBoxLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLPriorBoxLayerKernel. */
 class CLPriorBoxLayer : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLQLSTMLayer.h b/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
index 6e537680ee..a8f9221b3d 100644
--- a/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
@@ -24,9 +24,6 @@
 #ifndef ARM_COMPUTE_CLQLSTMLAYER_H
 #define ARM_COMPUTE_CLQLSTMLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLCopyKernel.h"
-#include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h"
-#include "arm_compute/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h"
@@ -40,7 +37,12 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
+class CLCopyKernel;
 class ICLTensor;
+class CLGEMMLowpMatrixAReductionKernel;
+class CLQLSTMLayerNormalizationKernel;
+class ITensorInfo;
 
 /** Basic function to run @ref CLQLSTMLayer
  *
@@ -68,6 +70,8 @@ public:
     CLQLSTMLayer &operator=(const CLQLSTMLayer &) = delete;
     /** Default move assignment operator */
     CLQLSTMLayer &operator=(CLQLSTMLayer &&) = default;
+    /** Default destructor */
+    ~CLQLSTMLayer();
     /** Initialize function's tensors.
      *
      * @param[in]  input                       Source tensor. Input is a 2D tensor with dimensions [input_size, batch_size]. Data types supported: QASYMM8_SIGNED.
@@ -285,72 +289,72 @@ private:
     };
 
     // Functions used
-    CLTranspose                      _transpose_input_to_forget_weights{};
-    CLTranspose                      _transpose_input_to_cell_weights{};
-    CLTranspose                      _transpose_input_to_output_weights{};
-    CLTranspose                      _transpose_input_to_input_weights{};
-    CLTranspose                      _transpose_recurrent_to_forget_weights{};
-    CLTranspose                      _transpose_recurrent_to_cell_weights{};
-    CLTranspose                      _transpose_recurrent_to_output_weights{};
-    CLTranspose                      _transpose_recurrent_to_input_weights{};
-    CLTranspose                      _transpose_projection_weights{};
-    CLGEMMLowpMatrixAReductionKernel _input_to_input_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _recurrent_to_input_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _input_to_forget_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _recurrent_to_forget_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _input_to_cell_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _recurrent_to_cell_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _input_to_output_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _recurrent_to_output_reduction{};
-    CLGEMMLowpMatrixAReductionKernel _projection_reduction{};
-    CLArithmeticAddition             _projection_bias_add{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_input_to_forget{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_recurrent_to_forget{};
-    CLPixelWiseMultiplication        _pixelwise_mul_cell_to_forget{};
-    CLGEMMLowpOutputStage            _input_to_forget_outstage{};
-    CLGEMMLowpOutputStage            _recurrent_to_forget_outstage{};
-    CLGEMMLowpOutputStage            _cell_to_forget_outstage{};
-    CLArithmeticAddition             _accumulate_input_recurrent_forget{};
-    CLArithmeticAddition             _accumulate_cell_forget{};
-    CLActivationLayer                _forget_gate_sigmoid{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_input_to_cell{};
-    CLGEMMLowpOutputStage            _input_to_cell_outstage{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_recurrent_to_cell{};
-    CLGEMMLowpOutputStage            _recurrent_to_cell_outstage{};
-    CLArithmeticAddition             _accumulate_input_recurrent_modulation{};
-    CLActivationLayer                _cell_gate_tanh{};
-    CLArithmeticSubtraction          _input_gate_sub{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_input_to_input{};
-    CLGEMMLowpOutputStage            _input_to_input_outstage{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_recurrent_to_input{};
-    CLGEMMLowpOutputStage            _recurrent_to_input_outstage{};
-    CLArithmeticAddition             _accumulate_input_recurrent_input{};
-    CLPixelWiseMultiplication        _pixelwise_mul_cell_to_input{};
-    CLGEMMLowpOutputStage            _cell_to_input_outstage{};
-    CLArithmeticAddition             _accumulate_cell_input{};
-    CLActivationLayer                _input_gate_sigmoid{};
-    CLPixelWiseMultiplication        _pixelwise_mul_forget_cell{};
-    CLPixelWiseMultiplication        _pixelwise_mul_input_cell{};
-    CLArithmeticAddition             _add_forget_cell{};
-    CLActivationLayer                _cell_clip{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_input_to_output{};
-    CLGEMMLowpOutputStage            _input_to_output_outstage{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_recurrent_to_output{};
-    CLGEMMLowpOutputStage            _recurrent_to_output_outstage{};
-    CLArithmeticAddition             _accumulate_input_recurrent_output{};
-    CLPixelWiseMultiplication        _pixelwise_mul_cell_to_output{};
-    CLGEMMLowpOutputStage            _cell_to_output_outstage{};
-    CLArithmeticAddition             _accumulate_cell_to_output{};
-    CLActivationLayer                _output_gate_sigmoid{};
-    CLActivationLayer                _hidden_tanh{};
-    CLPixelWiseMultiplication        _pixelwise_mul_hidden{};
-    CLGEMMLowpOutputStage            _hidden_outstage{};
-    CLGEMMLowpMatrixMultiplyCore     _mm_projection{};
-    CLGEMMLowpOutputStage            _projection_outstage{};
-    CLArithmeticAddition             _accumulate_projection{};
-    CLActivationLayer                _projection_clip{};
-    std::array<CLQLSTMLayerNormalizationKernel, _layer_norm_count> _layer_norms{ {} };
-    CLCopyKernel _copy_output{};
+    CLTranspose                                       _transpose_input_to_forget_weights{};
+    CLTranspose                                       _transpose_input_to_cell_weights{};
+    CLTranspose                                       _transpose_input_to_output_weights{};
+    CLTranspose                                       _transpose_input_to_input_weights{};
+    CLTranspose                                       _transpose_recurrent_to_forget_weights{};
+    CLTranspose                                       _transpose_recurrent_to_cell_weights{};
+    CLTranspose                                       _transpose_recurrent_to_output_weights{};
+    CLTranspose                                       _transpose_recurrent_to_input_weights{};
+    CLTranspose                                       _transpose_projection_weights{};
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _input_to_input_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _recurrent_to_input_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _input_to_forget_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _recurrent_to_forget_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _input_to_cell_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _recurrent_to_cell_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _input_to_output_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _recurrent_to_output_reduction;
+    std::unique_ptr<CLGEMMLowpMatrixAReductionKernel> _projection_reduction;
+    CLArithmeticAddition                              _projection_bias_add{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_input_to_forget{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_recurrent_to_forget{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_cell_to_forget{};
+    CLGEMMLowpOutputStage                             _input_to_forget_outstage{};
+    CLGEMMLowpOutputStage                             _recurrent_to_forget_outstage{};
+    CLGEMMLowpOutputStage                             _cell_to_forget_outstage{};
+    CLArithmeticAddition                              _accumulate_input_recurrent_forget{};
+    CLArithmeticAddition                              _accumulate_cell_forget{};
+    CLActivationLayer                                 _forget_gate_sigmoid{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_input_to_cell{};
+    CLGEMMLowpOutputStage                             _input_to_cell_outstage{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_recurrent_to_cell{};
+    CLGEMMLowpOutputStage                             _recurrent_to_cell_outstage{};
+    CLArithmeticAddition                              _accumulate_input_recurrent_modulation{};
+    CLActivationLayer                                 _cell_gate_tanh{};
+    CLArithmeticSubtraction                           _input_gate_sub{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_input_to_input{};
+    CLGEMMLowpOutputStage                             _input_to_input_outstage{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_recurrent_to_input{};
+    CLGEMMLowpOutputStage                             _recurrent_to_input_outstage{};
+    CLArithmeticAddition                              _accumulate_input_recurrent_input{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_cell_to_input{};
+    CLGEMMLowpOutputStage                             _cell_to_input_outstage{};
+    CLArithmeticAddition                              _accumulate_cell_input{};
+    CLActivationLayer                                 _input_gate_sigmoid{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_forget_cell{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_input_cell{};
+    CLArithmeticAddition                              _add_forget_cell{};
+    CLActivationLayer                                 _cell_clip{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_input_to_output{};
+    CLGEMMLowpOutputStage                             _input_to_output_outstage{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_recurrent_to_output{};
+    CLGEMMLowpOutputStage                             _recurrent_to_output_outstage{};
+    CLArithmeticAddition                              _accumulate_input_recurrent_output{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_cell_to_output{};
+    CLGEMMLowpOutputStage                             _cell_to_output_outstage{};
+    CLArithmeticAddition                              _accumulate_cell_to_output{};
+    CLActivationLayer                                 _output_gate_sigmoid{};
+    CLActivationLayer                                 _hidden_tanh{};
+    CLPixelWiseMultiplication                         _pixelwise_mul_hidden{};
+    CLGEMMLowpOutputStage                             _hidden_outstage{};
+    CLGEMMLowpMatrixMultiplyCore                      _mm_projection{};
+    CLGEMMLowpOutputStage                             _projection_outstage{};
+    CLArithmeticAddition                              _accumulate_projection{};
+    CLActivationLayer                                 _projection_clip{};
+    std::array<std::unique_ptr<CLQLSTMLayerNormalizationKernel>, _layer_norm_count> _layer_norms;
+    std::unique_ptr<CLCopyKernel> _copy_output;
 
     TensorCopyKernel _projection_bias_copy{};
     TensorCopyKernel _projection_output_to_accumulate_copy{};
@@ -402,30 +406,11 @@ private:
 
     inline CLQLSTMLayerNormalizationKernel &get_layer_norm(LayerNormGate g)
     {
-        return _layer_norms[getGateIndex(g)];
+        return *_layer_norms[getGateIndex(g)];
     }
 
-    inline void configure_layer_norm(LayerNormGate g, const ICLTensor *in)
-    {
-        ARM_COMPUTE_ERROR_ON(!_has_layer_norm);
-
-        CLTensor *out = &get_layer_norm_output(g);
-        _memory_group.manage(out);
-        out->allocator()->init(*(in->info()));
-
-        get_layer_norm(g).configure(in, out, get_layer_norm_weight(g), get_layer_norm_bias(g));
-    }
-
-    inline static Status validate_layer_norm(const ITensorInfo &in, const ITensorInfo &weight, const ITensorInfo &bias)
-    {
-        // Output quantization scale will be different, but ignored here
-        // since it will be configured at configure() stage.
-        const TensorInfo out
-        {
-            in
-        };
-        return CLQLSTMLayerNormalizationKernel::validate(&in, &out, &weight, &bias);
-    }
+    inline void configure_layer_norm(LayerNormGate g, const ICLTensor *in);
+    inline static Status validate_layer_norm(const ITensorInfo &in, const ITensorInfo &weight, const ITensorInfo &bias);
 
     // Temporary tensors
     CLTensor _input_to_forget_weights_transposed{ nullptr };
diff --git a/arm_compute/runtime/CL/functions/CLQuantizationLayer.h b/arm_compute/runtime/CL/functions/CLQuantizationLayer.h
index e045adf5fd..a0a27c5cb4 100644
--- a/arm_compute/runtime/CL/functions/CLQuantizationLayer.h
+++ b/arm_compute/runtime/CL/functions/CLQuantizationLayer.h
@@ -24,11 +24,14 @@
 #ifndef ARM_COMPUTE_CLQUANTIZATIONLAYER_H
 #define ARM_COMPUTE_CLQUANTIZATIONLAYER_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to simulate a quantization layer. This function calls the following CL kernels:
  *
diff --git a/arm_compute/runtime/CL/functions/CLRNNLayer.h b/arm_compute/runtime/CL/functions/CLRNNLayer.h
index 9d1cb1a724..ff3fb5449b 100644
--- a/arm_compute/runtime/CL/functions/CLRNNLayer.h
+++ b/arm_compute/runtime/CL/functions/CLRNNLayer.h
@@ -24,15 +24,17 @@
 #ifndef ARM_COMPUTE_CLRNN_LAYER_H
 #define ARM_COMPUTE_CLRNN_LAYER_H
 
-#include "arm_compute/core/CL/kernels/CLCopyKernel.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h"
 #include "arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h"
 #include "arm_compute/runtime/CL/functions/CLGEMM.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCopyKernel;
 class ICLTensor;
 
 /** Basic function to run @ref CLRNNLayer */
@@ -41,6 +43,12 @@ class CLRNNLayer : public IFunction
 public:
     /** Default constructor */
     CLRNNLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLRNNLayer(const CLRNNLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLRNNLayer &operator=(const CLRNNLayer &) = delete;
+    /** Default destructor */
+    ~CLRNNLayer();
     /** Initialize the function
      *
      * @param[in]     input             Input is a 2-D tensor of shape [input_size, batch_size]. Data types supported: F16/F32
@@ -85,16 +93,16 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup           _memory_group;
-    CLGEMM                _gemm_state_f;
-    CLArithmeticAddition  _add_kernel;
-    CLActivationLayer     _activation;
-    CLFullyConnectedLayer _fully_connected_kernel;
-    CLCopyKernel          _copy_kernel;
-    CLTensor              _fully_connected_out;
-    CLTensor              _gemm_output;
-    CLTensor              _add_output;
-    bool                  _is_prepared;
+    MemoryGroup                   _memory_group;
+    CLGEMM                        _gemm_state_f;
+    CLArithmeticAddition          _add_kernel;
+    CLActivationLayer             _activation;
+    CLFullyConnectedLayer         _fully_connected_kernel;
+    std::unique_ptr<CLCopyKernel> _copy_kernel;
+    CLTensor                      _fully_connected_out;
+    CLTensor                      _gemm_output;
+    CLTensor                      _add_output;
+    bool                          _is_prepared;
 };
 }
 #endif /* ARM_COMPUTE_CLRNN_LAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLROIAlignLayer.h b/arm_compute/runtime/CL/functions/CLROIAlignLayer.h
index 2e78f16d6b..b4cd5560ef 100644
--- a/arm_compute/runtime/CL/functions/CLROIAlignLayer.h
+++ b/arm_compute/runtime/CL/functions/CLROIAlignLayer.h
@@ -25,12 +25,14 @@
 #define ARM_COMPUTE_CLROIALIGNLAYER_H
 
 #include "arm_compute/core/CL/ICLArray.h"
-#include "arm_compute/core/CL/kernels/CLROIPoolingLayerKernel.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ROIPoolingLayerInfo;
+class ITensorInfo;
 
 /** Basic function to run @ref CLROIAlignLayerKernel.
  *
diff --git a/arm_compute/runtime/CL/functions/CLROIPoolingLayer.h b/arm_compute/runtime/CL/functions/CLROIPoolingLayer.h
index 30139274be..836575ef68 100644
--- a/arm_compute/runtime/CL/functions/CLROIPoolingLayer.h
+++ b/arm_compute/runtime/CL/functions/CLROIPoolingLayer.h
@@ -24,14 +24,14 @@
 #ifndef ARM_COMPUTE_CLROIPOOLINGLAYER_H
 #define ARM_COMPUTE_CLROIPOOLINGLAYER_H
 
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
-
 #include "arm_compute/core/CL/ICLArray.h"
-#include "arm_compute/core/CL/kernels/CLROIPoolingLayerKernel.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ROIPoolingLayerInfo;
 
 /** Basic function to run @ref CLROIPoolingLayerKernel.
  *
diff --git a/arm_compute/runtime/CL/functions/CLRange.h b/arm_compute/runtime/CL/functions/CLRange.h
index a86cfb605d..e11e740861 100644
--- a/arm_compute/runtime/CL/functions/CLRange.h
+++ b/arm_compute/runtime/CL/functions/CLRange.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLRangeKernel
  *
diff --git a/arm_compute/runtime/CL/functions/CLReductionOperation.h b/arm_compute/runtime/CL/functions/CLReductionOperation.h
index 5d050d71d6..3fbcee6c21 100644
--- a/arm_compute/runtime/CL/functions/CLReductionOperation.h
+++ b/arm_compute/runtime/CL/functions/CLReductionOperation.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLREDUCTIONOPERATION_H
 #define ARM_COMPUTE_CLREDUCTIONOPERATION_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLReductionOperationKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -37,6 +35,9 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLReductionOperationKernel;
 class ICLTensor;
 
 /** Perform reduction operation.
@@ -49,6 +50,16 @@ public:
      * @param[in] memory_manager (Optional) Memory manager.
      */
     CLReductionOperation(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Default Destructor */
+    ~CLReductionOperation();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLReductionOperation(const CLReductionOperation &) = delete;
+    /** Default move constructor */
+    CLReductionOperation(CLReductionOperation &&) = default;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLReductionOperation &operator=(const CLReductionOperation &) = delete;
+    /** Default move assignment operator */
+    CLReductionOperation &operator=(CLReductionOperation &&) = default;
 
     /** Set the input and output tensors.
      *
@@ -88,15 +99,15 @@ public:
 private:
     ICLTensor *configure_intermediate_result_vector(ICLTensor *input, ICLTensor *output);
 
-    MemoryGroup                             _memory_group;
-    std::vector<CLTensor>                   _results_vector;
-    std::vector<CLReductionOperationKernel> _reduction_kernels_vector;
-    std::vector<CLFillBorderKernel>         _border_handlers_vector;
-    CLReshapeLayer                          _reshape;
-    unsigned int                            _num_of_stages;
-    unsigned int                            _reduction_axis;
-    bool                                    _is_serial;
-    bool                                    _is_reshape_required;
+    MemoryGroup                                              _memory_group;
+    std::vector<CLTensor>                                    _results_vector;
+    std::vector<std::unique_ptr<CLReductionOperationKernel>> _reduction_kernels_vector;
+    std::vector<std::unique_ptr<CLFillBorderKernel>>         _border_handlers_vector;
+    CLReshapeLayer                                           _reshape;
+    unsigned int                                             _num_of_stages;
+    unsigned int                                             _reduction_axis;
+    bool                                                     _is_serial;
+    bool                                                     _is_reshape_required;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLREDUCTIONOPERATION_H */
\ No newline at end of file
diff --git a/arm_compute/runtime/CL/functions/CLRemap.h b/arm_compute/runtime/CL/functions/CLRemap.h
index 5b110d58f4..bf5d348b3b 100644
--- a/arm_compute/runtime/CL/functions/CLRemap.h
+++ b/arm_compute/runtime/CL/functions/CLRemap.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute remap. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLReorgLayer.h b/arm_compute/runtime/CL/functions/CLReorgLayer.h
index a7287ce266..0840fd13fd 100644
--- a/arm_compute/runtime/CL/functions/CLReorgLayer.h
+++ b/arm_compute/runtime/CL/functions/CLReorgLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 class CLReorgLayer : public ICLSimpleFunction
 {
diff --git a/arm_compute/runtime/CL/functions/CLReshapeLayer.h b/arm_compute/runtime/CL/functions/CLReshapeLayer.h
index 7fc6c3b864..b4d52ec8cf 100644
--- a/arm_compute/runtime/CL/functions/CLReshapeLayer.h
+++ b/arm_compute/runtime/CL/functions/CLReshapeLayer.h
@@ -29,7 +29,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLReshapeLayerKernel */
 class CLReshapeLayer : public IFunction
diff --git a/arm_compute/runtime/CL/functions/CLReverse.h b/arm_compute/runtime/CL/functions/CLReverse.h
index 6b140920e9..81fa04b1f5 100644
--- a/arm_compute/runtime/CL/functions/CLReverse.h
+++ b/arm_compute/runtime/CL/functions/CLReverse.h
@@ -24,11 +24,14 @@
 #ifndef ARM_COMPUTE_CLREVERSE_H
 #define ARM_COMPUTE_CLREVERSE_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLReverseKernel */
 class CLReverse : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLScale.h b/arm_compute/runtime/CL/functions/CLScale.h
index d776e83035..360d63ea22 100644
--- a/arm_compute/runtime/CL/functions/CLScale.h
+++ b/arm_compute/runtime/CL/functions/CLScale.h
@@ -32,7 +32,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLScaleKernel */
 class CLScale : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLScharr3x3.h b/arm_compute/runtime/CL/functions/CLScharr3x3.h
index 3892874f35..19c860f39b 100644
--- a/arm_compute/runtime/CL/functions/CLScharr3x3.h
+++ b/arm_compute/runtime/CL/functions/CLScharr3x3.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute scharr 3x3 filter. This function calls the following OpenCL kernels:
diff --git a/arm_compute/runtime/CL/functions/CLSelect.h b/arm_compute/runtime/CL/functions/CLSelect.h
index a1af922303..7fd52312fb 100644
--- a/arm_compute/runtime/CL/functions/CLSelect.h
+++ b/arm_compute/runtime/CL/functions/CLSelect.h
@@ -24,14 +24,15 @@
 #ifndef ARM_COMPUTE_CLSELECT_H
 #define ARM_COMPUTE_CLSELECT_H
 
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
-
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLSelect */
 class CLSelect : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLSlice.h b/arm_compute/runtime/CL/functions/CLSlice.h
index 23c398cb41..f17e77236d 100644
--- a/arm_compute/runtime/CL/functions/CLSlice.h
+++ b/arm_compute/runtime/CL/functions/CLSlice.h
@@ -31,6 +31,8 @@ namespace arm_compute
 {
 // Forward Declarations
 class ICLTensor;
+class CLCompileContext;
+class ITensorInfo;
 
 namespace experimental
 {
diff --git a/arm_compute/runtime/CL/functions/CLSobel3x3.h b/arm_compute/runtime/CL/functions/CLSobel3x3.h
index 25d4ed6895..492900da11 100644
--- a/arm_compute/runtime/CL/functions/CLSobel3x3.h
+++ b/arm_compute/runtime/CL/functions/CLSobel3x3.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to execute sobel 3x3 filter. This function calls the following OpenCL kernels:
@@ -42,6 +43,14 @@ class ICLTensor;
 class CLSobel3x3 : public ICLSimpleFunction
 {
 public:
+    /** Default Constructor */
+    CLSobel3x3() = default;
+    /** Prevent instances of this class from being copied */
+    CLSobel3x3(const CLSobel3x3 &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLSobel3x3 &operator=(const CLSobel3x3 &) = delete;
+    /** Default destructor */
+    ~CLSobel3x3();
     /** Initialise the function's source, destinations and border mode.
      *
      * @note At least one of output_x or output_y must be not NULL.
diff --git a/arm_compute/runtime/CL/functions/CLSobel5x5.h b/arm_compute/runtime/CL/functions/CLSobel5x5.h
index 1f91c46f7f..a00fdd72b8 100644
--- a/arm_compute/runtime/CL/functions/CLSobel5x5.h
+++ b/arm_compute/runtime/CL/functions/CLSobel5x5.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLSOBEL5X5_H
 #define ARM_COMPUTE_CLSOBEL5X5_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLSobel5x5Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -37,6 +35,10 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLSobel5x5HorKernel;
+class CLSobel5x5VertKernel;
 class ICLTensor;
 
 /** Basic function to execute sobel 5x5 filter. This function calls the following OpenCL kernels:
@@ -54,6 +56,12 @@ public:
      * @param[in] memory_manager (Optional) Memory manager.
      */
     CLSobel5x5(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLSobel5x5(const CLSobel5x5 &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLSobel5x5 &operator=(const CLSobel5x5 &) = delete;
+    /** Default destructor */
+    ~CLSobel5x5();
     /** Initialise the function's source, destinations and border mode.
      *
      * @note At least one of output_x or output_y must be not NULL.
@@ -82,12 +90,12 @@ public:
     void run() override;
 
 protected:
-    MemoryGroup          _memory_group;   /**< Function's memory group */
-    CLSobel5x5HorKernel  _sobel_hor;      /**< Sobel Horizontal 5x5 kernel */
-    CLSobel5x5VertKernel _sobel_vert;     /**< Sobel Vertical 5x5 kernel */
-    CLFillBorderKernel   _border_handler; /**< Kernel to handle image borders */
-    CLImage              _tmp_x;          /**< Temporary buffer for Sobel X */
-    CLImage              _tmp_y;          /**< Temporary buffer for Sobel Y */
+    MemoryGroup                           _memory_group;   /**< Function's memory group */
+    std::unique_ptr<CLSobel5x5HorKernel>  _sobel_hor;      /**< Sobel Horizontal 5x5 kernel */
+    std::unique_ptr<CLSobel5x5VertKernel> _sobel_vert;     /**< Sobel Vertical 5x5 kernel */
+    std::unique_ptr<CLFillBorderKernel>   _border_handler; /**< Kernel to handle image borders */
+    CLImage                               _tmp_x;          /**< Temporary buffer for Sobel X */
+    CLImage                               _tmp_y;          /**< Temporary buffer for Sobel Y */
 };
 }
 #endif /*ARM_COMPUTE_CLSOBEL5X5_H */
diff --git a/arm_compute/runtime/CL/functions/CLSobel7x7.h b/arm_compute/runtime/CL/functions/CLSobel7x7.h
index 91daf64c29..01a863b11b 100644
--- a/arm_compute/runtime/CL/functions/CLSobel7x7.h
+++ b/arm_compute/runtime/CL/functions/CLSobel7x7.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLSOBEL7X7_H
 #define ARM_COMPUTE_CLSOBEL7X7_H
 
-#include "arm_compute/core/CL/kernels/CLFillBorderKernel.h"
-#include "arm_compute/core/CL/kernels/CLSobel7x7Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -37,6 +35,10 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLFillBorderKernel;
+class CLSobel7x7HorKernel;
+class CLSobel7x7VertKernel;
 class ICLTensor;
 
 /** Basic function to execute sobel 7x7 filter. This function calls the following OpenCL kernels:
@@ -54,6 +56,12 @@ public:
      * @param[in] memory_manager (Optional) Memory manager.
      */
     CLSobel7x7(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLSobel7x7(const CLSobel7x7 &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLSobel7x7 &operator=(const CLSobel7x7 &) = delete;
+    /** Default destructor */
+    ~CLSobel7x7();
     /** Initialise the function's source, destinations and border mode.
      *
      * @note At least one of output_x or output_y must be not NULL.
@@ -82,12 +90,12 @@ public:
     void run() override;
 
 protected:
-    MemoryGroup          _memory_group;   /**< Function's memory group */
-    CLSobel7x7HorKernel  _sobel_hor;      /**< Sobel Horizontal 7x7 kernel */
-    CLSobel7x7VertKernel _sobel_vert;     /**< Sobel Vertical 7x7 kernel */
-    CLFillBorderKernel   _border_handler; /**< Kernel to handle image borders */
-    CLImage              _tmp_x;          /**< Temporary buffer for Sobel X */
-    CLImage              _tmp_y;          /**< Temporary buffer for Sobel Y */
+    MemoryGroup                           _memory_group;   /**< Function's memory group */
+    std::unique_ptr<CLSobel7x7HorKernel>  _sobel_hor;      /**< Sobel Horizontal 7x7 kernel */
+    std::unique_ptr<CLSobel7x7VertKernel> _sobel_vert;     /**< Sobel Vertical 7x7 kernel */
+    std::unique_ptr<CLFillBorderKernel>   _border_handler; /**< Kernel to handle image borders */
+    CLImage                               _tmp_x;          /**< Temporary buffer for Sobel X */
+    CLImage                               _tmp_y;          /**< Temporary buffer for Sobel Y */
 };
 }
 #endif /*ARM_COMPUTE_CLSOBEL7X7_H */
diff --git a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h
index fd71f3ed4d..ab10a64de4 100644
--- a/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h
+++ b/arm_compute/runtime/CL/functions/CLSoftmaxLayer.h
@@ -24,7 +24,6 @@
 #ifndef ARM_COMPUTE_CLSOFTMAXLAYER_H
 #define ARM_COMPUTE_CLSOFTMAXLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLSoftmaxLayerKernel.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLPermute.h"
 #include "arm_compute/runtime/IFunction.h"
@@ -35,7 +34,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLLogits1DMaxShiftExpSumKernel;
+class CLLogits1DNormKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to compute a SoftmaxLayer.
  *
@@ -57,6 +60,16 @@ class CLSoftmaxLayerGeneric : public IFunction
 public:
     /** Constructor */
     CLSoftmaxLayerGeneric(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
+    /** Prevent instances of this class from being copied */
+    CLSoftmaxLayerGeneric(const CLSoftmaxLayerGeneric &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLSoftmaxLayerGeneric &operator=(const CLSoftmaxLayerGeneric &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLSoftmaxLayerGeneric(CLSoftmaxLayerGeneric &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLSoftmaxLayerGeneric &operator=(CLSoftmaxLayerGeneric &&) = delete;
+    /** Default destructor */
+    ~CLSoftmaxLayerGeneric();
     /** Set the input and output tensors.
      *
      * @param[in]  input  Source tensor. Data types supported: QASYMM8/QASYMM8_SIGNED/F16/F32 for Softmax and F16/F32 for Log Softmax
@@ -92,17 +105,17 @@ public:
     void run() override;
 
 private:
-    MemoryGroup                    _memory_group;
-    CLPermute                      _permute_input;
-    CLPermute                      _permute_output;
-    CLLogits1DMaxShiftExpSumKernel _max_shift_exp_sum_kernel;
-    CLLogits1DNormKernel           _norm_kernel;
-    CLTensor                       _max;
-    CLTensor                       _sum;
-    CLTensor                       _tmp;
-    CLTensor                       _input_permuted;
-    CLTensor                       _output_permuted;
-    bool                           _needs_permute;
+    MemoryGroup                                     _memory_group;
+    CLPermute                                       _permute_input;
+    CLPermute                                       _permute_output;
+    std::unique_ptr<CLLogits1DMaxShiftExpSumKernel> _max_shift_exp_sum_kernel;
+    std::unique_ptr<CLLogits1DNormKernel>           _norm_kernel;
+    CLTensor                                        _max;
+    CLTensor                                        _sum;
+    CLTensor                                        _tmp;
+    CLTensor                                        _input_permuted;
+    CLTensor                                        _output_permuted;
+    bool                                            _needs_permute;
 };
 
 using CLSoftmaxLayer    = CLSoftmaxLayerGeneric<false>;
diff --git a/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
index c6f7f11079..1611aa8ed4 100644
--- a/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
+++ b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
@@ -24,16 +24,19 @@
 #ifndef ARM_COMPUTE_CLSPACETOBATCHLAYER_H
 #define ARM_COMPUTE_CLSPACETOBATCHLAYER_H
 
-#include "arm_compute/runtime/IFunction.h"
-
-#include "arm_compute/core/CL/kernels/CLMemsetKernel.h"
-#include "arm_compute/core/CL/kernels/CLSpaceToBatchLayerKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/IFunction.h"
+
+#include <memory>
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLMemsetKernel;
+class CLSpaceToBatchLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to spatial divide a tensor. This function calls the following OpenCL kernels/functions:
  *
@@ -54,7 +57,7 @@ public:
     /** Allow instances of this class to be moved */
     CLSpaceToBatchLayer &operator=(CLSpaceToBatchLayer &&) = default;
     /** Default destructor */
-    virtual ~CLSpaceToBatchLayer() = default;
+    ~CLSpaceToBatchLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input       Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -121,9 +124,9 @@ public:
     void run() override;
 
 private:
-    CLSpaceToBatchLayerKernel _space_to_batch_kernel; /**< SpaceToBatch kernel to run */
-    CLMemsetKernel            _memset_kernel;         /**< Memset kernel to run */
-    bool                      _has_padding;           /**< Flag to check if the output has padding */
+    std::unique_ptr<CLSpaceToBatchLayerKernel> _space_to_batch_kernel; /**< SpaceToBatch kernel to run */
+    std::unique_ptr<CLMemsetKernel>            _memset_kernel;         /**< Memset kernel to run */
+    bool                                       _has_padding;           /**< Flag to check if the output has padding */
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLSPACETOBATCHLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLSpaceToDepthLayer.h b/arm_compute/runtime/CL/functions/CLSpaceToDepthLayer.h
index 24830cf4d3..9e476fe7bd 100644
--- a/arm_compute/runtime/CL/functions/CLSpaceToDepthLayer.h
+++ b/arm_compute/runtime/CL/functions/CLSpaceToDepthLayer.h
@@ -24,14 +24,17 @@
 #ifndef ARM_COMPUTE_CLSPACETODEPTHLAYER_H
 #define ARM_COMPUTE_CLSPACETODEPTHLAYER_H
 
+#include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLSpaceToDepthLayerKernel.h"
-#include "arm_compute/core/Types.h"
+#include <memory>
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLSpaceToDepthLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLSpaceToDepthLayerKernel. */
 class CLSpaceToDepthLayer : public IFunction
@@ -39,6 +42,16 @@ class CLSpaceToDepthLayer : public IFunction
 public:
     /** Default constructor */
     CLSpaceToDepthLayer();
+    /** Prevent instances of this class from being copied */
+    CLSpaceToDepthLayer(const CLSpaceToDepthLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLSpaceToDepthLayer &operator=(const CLSpaceToDepthLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLSpaceToDepthLayer(CLSpaceToDepthLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLSpaceToDepthLayer &operator=(CLSpaceToDepthLayer &&) = delete;
+    /** Default destructor */
+    ~CLSpaceToDepthLayer();
     /** Set the input and output tensors.
      *
      * @param[in]  input       Tensor input. Supported tensor rank: 4. Data types supported: All.
@@ -68,7 +81,7 @@ public:
     void run() override;
 
 private:
-    CLSpaceToDepthLayerKernel _space_to_depth_kernel; /**< CLSpaceToDepthLayerKernel to run */
+    std::unique_ptr<CLSpaceToDepthLayerKernel> _space_to_depth_kernel; /**< CLSpaceToDepthLayerKernel to run */
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLSPACETODEPTHLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLStackLayer.h b/arm_compute/runtime/CL/functions/CLStackLayer.h
index 95875962c8..3861fd299a 100644
--- a/arm_compute/runtime/CL/functions/CLStackLayer.h
+++ b/arm_compute/runtime/CL/functions/CLStackLayer.h
@@ -27,14 +27,15 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLStackLayerKernel.h"
-
 #include <memory>
 #include <vector>
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLStackLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to stack tensors along an axis. This function calls the following kernel:
  *
@@ -46,6 +47,16 @@ class CLStackLayer : public IFunction
 public:
     /** Default constructor */
     CLStackLayer();
+    /** Prevent instances of this class from being copied */
+    CLStackLayer(const CLStackLayer &) = delete;
+    /** Prevent instances of this class from being copied */
+    CLStackLayer &operator=(const CLStackLayer &) = delete;
+    /** Prevent instances of this class to be moved */
+    CLStackLayer(CLStackLayer &&) = delete;
+    /** Prevent instances of this class to be moved */
+    CLStackLayer &operator=(CLStackLayer &&) = delete;
+    /** Default destructor */
+    ~CLStackLayer();
     /** Initialise the kernel's inputs vector and output.
      *
      * @note Supported input tensor rank: up to 4
@@ -84,9 +95,9 @@ public:
     void run() override;
 
 private:
-    std::vector<ICLTensor *>        _input;
-    std::vector<CLStackLayerKernel> _stack_kernels;
-    unsigned int                    _num_inputs;
+    std::vector<ICLTensor *>                         _input;
+    std::vector<std::unique_ptr<CLStackLayerKernel>> _stack_kernels;
+    unsigned int                                     _num_inputs;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLSTACKLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLTableLookup.h b/arm_compute/runtime/CL/functions/CLTableLookup.h
index 32d4b7bdf9..ca59309548 100644
--- a/arm_compute/runtime/CL/functions/CLTableLookup.h
+++ b/arm_compute/runtime/CL/functions/CLTableLookup.h
@@ -28,6 +28,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 class ICLLut;
 
diff --git a/arm_compute/runtime/CL/functions/CLThreshold.h b/arm_compute/runtime/CL/functions/CLThreshold.h
index f3af122f0a..2c9213bd01 100644
--- a/arm_compute/runtime/CL/functions/CLThreshold.h
+++ b/arm_compute/runtime/CL/functions/CLThreshold.h
@@ -33,6 +33,7 @@
 namespace arm_compute
 {
 // Forward declarations
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLThresholdKernel */
diff --git a/arm_compute/runtime/CL/functions/CLTile.h b/arm_compute/runtime/CL/functions/CLTile.h
index d2f1e9730c..69743693ff 100644
--- a/arm_compute/runtime/CL/functions/CLTile.h
+++ b/arm_compute/runtime/CL/functions/CLTile.h
@@ -24,13 +24,14 @@
 #ifndef ARM_COMPUTE_CLTILE_H
 #define ARM_COMPUTE_CLTILE_H
 
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
-
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLTileKernel */
 class CLTile : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLTranspose.h b/arm_compute/runtime/CL/functions/CLTranspose.h
index 9ba7cafce4..2b7a03f23f 100644
--- a/arm_compute/runtime/CL/functions/CLTranspose.h
+++ b/arm_compute/runtime/CL/functions/CLTranspose.h
@@ -24,11 +24,14 @@
 #ifndef ARM_COMPUTE_CLTRANSPOSE_H
 #define ARM_COMPUTE_CLTRANSPOSE_H
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to transpose a matrix on OpenCL. This function calls the following OpenCL kernel:
  *
diff --git a/arm_compute/runtime/CL/functions/CLUpsampleLayer.h b/arm_compute/runtime/CL/functions/CLUpsampleLayer.h
index 07b4c8aecb..88b293069d 100644
--- a/arm_compute/runtime/CL/functions/CLUpsampleLayer.h
+++ b/arm_compute/runtime/CL/functions/CLUpsampleLayer.h
@@ -26,13 +26,17 @@
 
 #include "arm_compute/runtime/IFunction.h"
 
-#include "arm_compute/core/CL/kernels/CLUpsampleLayerKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
 
+#include <memory>
+
 namespace arm_compute
 {
+class CLCompileContext;
+class CLUpsampleLayerKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLUpsampleLayerKernel */
 class CLUpsampleLayer : public IFunction
@@ -49,7 +53,7 @@ public:
     /** Allow instances of this class to be moved */
     CLUpsampleLayer &operator=(CLUpsampleLayer &&) = default;
     /** Default destructor */
-    virtual ~CLUpsampleLayer() = default;
+    ~CLUpsampleLayer();
 
     /** Initialize the function's source, destination, interpolation type and border_mode.
      *
@@ -86,8 +90,8 @@ public:
     void run() override;
 
 private:
-    CLUpsampleLayerKernel _upsample;
-    ICLTensor            *_output;
+    std::unique_ptr<CLUpsampleLayerKernel> _upsample;
+    ICLTensor                             *_output;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLUPSAMPLELAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLWarpAffine.h b/arm_compute/runtime/CL/functions/CLWarpAffine.h
index eb7c05be84..153e9bfdfc 100644
--- a/arm_compute/runtime/CL/functions/CLWarpAffine.h
+++ b/arm_compute/runtime/CL/functions/CLWarpAffine.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLWarpAffineKernel for AFFINE transformation */
diff --git a/arm_compute/runtime/CL/functions/CLWarpPerspective.h b/arm_compute/runtime/CL/functions/CLWarpPerspective.h
index 2a1f78093e..5c8b5425a4 100644
--- a/arm_compute/runtime/CL/functions/CLWarpPerspective.h
+++ b/arm_compute/runtime/CL/functions/CLWarpPerspective.h
@@ -31,6 +31,7 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
 
 /** Basic function to run @ref CLWarpPerspectiveKernel for PERSPECTIVE transformation */
diff --git a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h
index 602f644230..9ced69c1bb 100644
--- a/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLWinogradConvolutionLayer.h
@@ -24,8 +24,6 @@
 #ifndef ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H
 #define ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H
 
-#include "arm_compute/core/CL/kernels/CLWinogradFilterTransformKernel.h"
-#include "arm_compute/core/CL/kernels/CLWinogradOutputTransformKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/functions/CLGEMM.h"
 #include "arm_compute/runtime/CL/functions/CLWinogradInputTransform.h"
@@ -33,7 +31,11 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
+class CLWinogradFilterTransformKernel;
+class CLWinogradOutputTransformKernel;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute Winograd-based convolution on OpenCL. This function calls the following OpenCL functions/kernels:
  *
@@ -56,6 +58,8 @@ public:
     CLWinogradConvolutionLayer &operator=(const CLWinogradConvolutionLayer &) = delete;
     /** Default move assignment operator */
     CLWinogradConvolutionLayer &operator=(CLWinogradConvolutionLayer &&) = default;
+    /** Default destructor */
+    ~CLWinogradConvolutionLayer();
     /** Set the input and output tensors.
      *
      * @note: This function only works with 3x3,3x1,1x3,5x5,5x1,1x5,7x1 and 1x7 kernels along with unit strides for both NCHW and NHWC data layout
@@ -122,16 +126,16 @@ public:
     void prepare() override;
 
 private:
-    MemoryGroup                     _memory_group;
-    CLGEMM                          _batched_mm;
-    CLWinogradInputTransform        _input_transform;
-    CLWinogradFilterTransformKernel _filter_transform;
-    CLWinogradOutputTransformKernel _output_transform;
-    CLTensor                        _input0;
-    CLTensor                        _input1;
-    CLTensor                        _batched_mm_output;
-    const ICLTensor                *_original_weights;
-    bool                            _is_prepared;
+    MemoryGroup                                      _memory_group;
+    CLGEMM                                           _batched_mm;
+    CLWinogradInputTransform                         _input_transform;
+    std::unique_ptr<CLWinogradFilterTransformKernel> _filter_transform;
+    std::unique_ptr<CLWinogradOutputTransformKernel> _output_transform;
+    CLTensor                                         _input0;
+    CLTensor                                         _input1;
+    CLTensor                                         _batched_mm_output;
+    const ICLTensor                                 *_original_weights;
+    bool                                             _is_prepared;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CLWINOGRADCONVOLUTIONLAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h b/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h
index 351f88012f..8cd809cc1f 100644
--- a/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h
+++ b/arm_compute/runtime/CL/functions/CLWinogradInputTransform.h
@@ -31,7 +31,9 @@
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to execute a @ref CLWinogradInputTransformKernel. */
 class CLWinogradInputTransform : public ICLSimpleFunction
diff --git a/arm_compute/runtime/CL/functions/CLYOLOLayer.h b/arm_compute/runtime/CL/functions/CLYOLOLayer.h
index 3e403f44bd..48ee4ea4f7 100644
--- a/arm_compute/runtime/CL/functions/CLYOLOLayer.h
+++ b/arm_compute/runtime/CL/functions/CLYOLOLayer.h
@@ -24,13 +24,14 @@
 #ifndef ARM_COMPUTE_CLYOLOLAYER_H
 #define ARM_COMPUTE_CLYOLOLAYER_H
 
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
-
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
 /** Basic function to run @ref CLYOLOLayerKernel that performs a partial activation on the input
  *
diff --git a/arm_compute/runtime/IOperator.h b/arm_compute/runtime/IOperator.h
index 0097383115..fd285160e9 100644
--- a/arm_compute/runtime/IOperator.h
+++ b/arm_compute/runtime/IOperator.h
@@ -24,14 +24,13 @@
 #ifndef ARM_COMPUTE_IOPERATOR_H
 #define ARM_COMPUTE_IOPERATOR_H
 
-#include "arm_compute/core/ITensorPack.h"
 #include "arm_compute/core/experimental/Types.h"
-#include "arm_compute/runtime/IOperator.h"
 #include "arm_compute/runtime/IRuntimeContext.h"
 #include "arm_compute/runtime/Types.h"
 
 namespace arm_compute
 {
+class ITensorPack;
 namespace experimental
 {
 /** Base class for all async functions */