Make memset/copy functions state-less

Port following functions:
- CLCopy
- CLFill
- CLPermute
- CLReshapeLayer
- CLCropResize

Resolves: COMPMID-4002

Signed-off-by: Sheri Zhang <sheri.zhang@arm.com>
Change-Id: I8392aa515aaeb5b44dab6122be6a795d08376d5f
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5003
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

58 files changed, 1582 insertions, 717 deletions

diff --git a/Android.bp b/Android.bp
index 5653fc8a63..94a4e6c13a 100644
--- a/Android.bp
+++ b/Android.bp
@@ -95,8 +95,6 @@ cc_library_static {
         "src/core/CL/kernels/CLComparisonKernel.cpp",
         "src/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.cpp",
         "src/core/CL/kernels/CLConvolutionKernel.cpp",
-        "src/core/CL/kernels/CLCopyKernel.cpp",
-        "src/core/CL/kernels/CLCropKernel.cpp",
         "src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.cpp",
         "src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.cpp",
         "src/core/CL/kernels/CLDepthConvertLayerKernel.cpp",
@@ -150,7 +148,6 @@ cc_library_static {
         "src/core/CL/kernels/CLMeanStdDevKernel.cpp",
         "src/core/CL/kernels/CLMeanStdDevNormalizationKernel.cpp",
         "src/core/CL/kernels/CLMedian3x3Kernel.cpp",
-        "src/core/CL/kernels/CLMemsetKernel.cpp",
         "src/core/CL/kernels/CLMinMaxLayerKernel.cpp",
         "src/core/CL/kernels/CLMinMaxLocationKernel.cpp",
         "src/core/CL/kernels/CLNonLinearFilterKernel.cpp",
@@ -158,7 +155,6 @@ cc_library_static {
         "src/core/CL/kernels/CLNormalizationLayerKernel.cpp",
         "src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.cpp",
         "src/core/CL/kernels/CLPadLayerKernel.cpp",
-        "src/core/CL/kernels/CLPermuteKernel.cpp",
         "src/core/CL/kernels/CLPixelWiseMultiplicationKernel.cpp",
         "src/core/CL/kernels/CLPriorBoxLayerKernel.cpp",
         "src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.cpp",
@@ -169,7 +165,6 @@ cc_library_static {
         "src/core/CL/kernels/CLReductionOperationKernel.cpp",
         "src/core/CL/kernels/CLRemapKernel.cpp",
         "src/core/CL/kernels/CLReorgLayerKernel.cpp",
-        "src/core/CL/kernels/CLReshapeLayerKernel.cpp",
         "src/core/CL/kernels/CLReverseKernel.cpp",
         "src/core/CL/kernels/CLScaleKernel.cpp",
         "src/core/CL/kernels/CLScharr3x3Kernel.cpp",
@@ -437,12 +432,17 @@ cc_library_static {
         "src/core/cpu/kernels/sub/neon/qsymm16.cpp",
         "src/core/gpu/cl/kernels/ClActivationKernel.cpp",
         "src/core/gpu/cl/kernels/ClBatchConcatenateKernel.cpp",
+        "src/core/gpu/cl/kernels/ClCopyKernel.cpp",
+        "src/core/gpu/cl/kernels/ClCropKernel.cpp",
         "src/core/gpu/cl/kernels/ClDepthConcatenateKernel.cpp",
         "src/core/gpu/cl/kernels/ClElementwiseKernel.cpp",
         "src/core/gpu/cl/kernels/ClElementwiseUnaryKernel.cpp",
+        "src/core/gpu/cl/kernels/ClFillKernel.cpp",
         "src/core/gpu/cl/kernels/ClFloorKernel.cpp",
         "src/core/gpu/cl/kernels/ClHeightConcatenateKernel.cpp",
+        "src/core/gpu/cl/kernels/ClPermuteKernel.cpp",
         "src/core/gpu/cl/kernels/ClPoolingKernel.cpp",
+        "src/core/gpu/cl/kernels/ClReshapeKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenate2TensorsKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenate4TensorsKernel.cpp",
         "src/core/gpu/cl/kernels/ClWidthConcatenateKernel.cpp",
@@ -505,6 +505,7 @@ cc_library_static {
         "src/runtime/CL/functions/CLConvolution.cpp",
         "src/runtime/CL/functions/CLConvolutionLayer.cpp",
         "src/runtime/CL/functions/CLCopy.cpp",
+        "src/runtime/CL/functions/CLCrop.cpp",
         "src/runtime/CL/functions/CLCropResize.cpp",
         "src/runtime/CL/functions/CLDeconvolutionLayer.cpp",
         "src/runtime/CL/functions/CLDeconvolutionLayerUpsample.cpp",
@@ -802,11 +803,16 @@ cc_library_static {
         "src/runtime/gpu/cl/operators/ClActivation.cpp",
         "src/runtime/gpu/cl/operators/ClAdd.cpp",
         "src/runtime/gpu/cl/operators/ClConcatenate.cpp",
+        "src/runtime/gpu/cl/operators/ClCopy.cpp",
+        "src/runtime/gpu/cl/operators/ClCrop.cpp",
         "src/runtime/gpu/cl/operators/ClElementwiseOperations.cpp",
         "src/runtime/gpu/cl/operators/ClElementwiseUnary.cpp",
+        "src/runtime/gpu/cl/operators/ClFill.cpp",
         "src/runtime/gpu/cl/operators/ClFloor.cpp",
         "src/runtime/gpu/cl/operators/ClLogicalNot.cpp",
+        "src/runtime/gpu/cl/operators/ClPermute.cpp",
         "src/runtime/gpu/cl/operators/ClPooling.cpp",
+        "src/runtime/gpu/cl/operators/ClReshape.cpp",
         "src/runtime/gpu/cl/operators/ClSub.cpp",
         "utils/CommonGraphOptions.cpp",
         "utils/GraphUtils.cpp",
diff --git a/arm_compute/runtime/CL/CLFunctions.h b/arm_compute/runtime/CL/CLFunctions.h
index 26c2670cbc..b2bdb9a3e7 100644
--- a/arm_compute/runtime/CL/CLFunctions.h
+++ b/arm_compute/runtime/CL/CLFunctions.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016-2020 Arm Limited.
+ * Copyright (c) 2016-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -49,6 +49,7 @@
 #include "arm_compute/runtime/CL/functions/CLConvolution.h"
 #include "arm_compute/runtime/CL/functions/CLConvolutionLayer.h"
 #include "arm_compute/runtime/CL/functions/CLCopy.h"
+#include "arm_compute/runtime/CL/functions/CLCrop.h"
 #include "arm_compute/runtime/CL/functions/CLCropResize.h"
 #include "arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h"
 #include "arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h"
diff --git a/arm_compute/runtime/CL/functions/CLCopy.h b/arm_compute/runtime/CL/functions/CLCopy.h
index f1a091df84..795a183e1f 100644
--- a/arm_compute/runtime/CL/functions/CLCopy.h
+++ b/arm_compute/runtime/CL/functions/CLCopy.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,9 +25,9 @@
 #define ARM_COMPUTE_CLCOPY_H
 
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
-
-#include <cstdint>
+#include "arm_compute/core/Window.h"
+#include "arm_compute/runtime/IFunction.h"
+#include <memory>
 
 namespace arm_compute
 {
@@ -35,32 +35,54 @@ class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 
-class CLCopy : public ICLSimpleFunction
+/** Basic function to run @ref opencl::kernels::ClCopyKernel */
+class CLCopy : public IFunction
 {
 public:
+    /** Constructor */
+    CLCopy();
+    /** Destructor */
+    ~CLCopy();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLCopy(const CLCopy &) = delete;
+    /** Default move constructor */
+    CLCopy(CLCopy &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLCopy &operator=(const CLCopy &) = delete;
+    /** Default move assignment operator */
+    CLCopy &operator=(CLCopy &&);
     /** Initialise the function's source and destination.
      *
-     * @param[in]  input  Source tensor. Data types supported: All.
-     * @param[out] output Output tensor. Data types supported: Same as @p input.
-     *
+     * @param[in]  input      Source tensor. Data types supported: All.
+     * @param[out] output     Output tensor. Data types supported: Same as @p input.
+     * @param[in]  dst_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
      */
-    void configure(ICLTensor *input, ICLTensor *output);
+    void configure(ICLTensor *input, ICLTensor *output, Window *dst_window = nullptr);
     /** Initialise the function's source and destination.
      *
      * @param[in]  compile_context The compile context to be used.
      * @param[in]  input           Source tensor. Data types supported: All.
      * @param[out] output          Output tensor. Data types supported: Same as @p input.
+     * @param[in]  dst_window      (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
      *
      */
-    void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output);
+    void configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, Window *dst_window = nullptr);
     /** Static function to check if given info will lead to a valid configuration of @ref CLCopy
      *
-     * @param[in] input  Source tensor. Data types supported: All.
-     * @param[in] output Output tensor. Data types supported: Same as @p input.
+     * @param[in] input      Source tensor. Data types supported: All.
+     * @param[in] output     Output tensor. Data types supported: Same as @p input.
+     * @param[in] dst_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
      *
      * @return a status
      */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, Window *dst_window = nullptr);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLCOPY_H */
diff --git a/src/core/CL/kernels/CLCropKernel.h b/arm_compute/runtime/CL/functions/CLCrop.h
index cbfada58ab..dc509b5b84 100644
--- a/src/core/CL/kernels/CLCropKernel.h
+++ b/arm_compute/runtime/CL/functions/CLCrop.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,31 +21,37 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_CLCROPKERNEL_H
-#define ARM_COMPUTE_CLCROPKERNEL_H
+#ifndef ARM_COMPUTE_CL_CROP_H
+#define ARM_COMPUTE_CL_CROP_H
 
 #include "arm_compute/core/Types.h"
-#include "src/core/CL/ICLKernel.h"
+#include "arm_compute/core/Window.h"
+#include "arm_compute/runtime/IFunction.h"
+#include <memory>
 
 namespace arm_compute
 {
+class CLCompileContext;
 class ICLTensor;
+class ITensorInfo;
 
-/** OpenCL kernel to perform a copy between two tensors */
-class CLCropKernel : public ICLKernel
+/** Basic function to run @ref opencl::kernels::ClCropKernel */
+class CLCrop : public IFunction
 {
 public:
-    /** Default constructor */
-    CLCropKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers). */
-    CLCropKernel(const CLCropKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers). */
-    CLCropKernel &operator=(const CLCropKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLCropKernel(CLCropKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLCropKernel &operator=(CLCropKernel &&) = default;
-    /** Configure kernel
+    /** Constructor */
+    CLCrop();
+    /** Destructor */
+    ~CLCrop();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLCrop(const CLCrop &) = delete;
+    /** Default move constructor */
+    CLCrop(CLCrop &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLCrop &operator=(const CLCrop &) = delete;
+    /** Default move assignment operator */
+    CLCrop &operator=(CLCrop &&);
+    /** Configure function
      *
      * @note Supported tensor rank: up to 4
      *
@@ -58,7 +64,7 @@ public:
      * @param[in]  output_window       Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
      */
     void configure(const ICLTensor *input, ICLTensor *output, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0, Window *output_window = nullptr);
-    /** Configure kernel
+    /** Configure function
      *
      * @note Supported tensor rank: up to 4
      *
@@ -90,14 +96,11 @@ public:
                            Window *output_window = nullptr);
 
     // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
+    void run() override;
 
 private:
-    const ICLTensor *_input;
-    ICLTensor       *_output;
-    Coordinates2D    _start;
-    uint32_t         _batch_index;
-    float            _extrapolation_value;
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
-#endif /*ARM_COMPUTE_CLCROPKERNEL_H */
+#endif /*ARM_COMPUTE_CL_CROP_H */
diff --git a/arm_compute/runtime/CL/functions/CLCropResize.h b/arm_compute/runtime/CL/functions/CLCropResize.h
index e781cfe61f..0dc3c48b32 100644
--- a/arm_compute/runtime/CL/functions/CLCropResize.h
+++ b/arm_compute/runtime/CL/functions/CLCropResize.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,6 +27,9 @@
 #include "arm_compute/core/CL/ICLTensor.h"
 
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLCopy.h"
+#include "arm_compute/runtime/CL/functions/CLCrop.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/CL/functions/CLScale.h"
 
 #include <cstdint>
@@ -36,8 +39,6 @@ namespace arm_compute
 {
 // Forward Declarations
 class CLCompileContext;
-class CLCopyKernel;
-class CLCropKernel;
 class ITensor;
 class ITensorInfo;
 
@@ -125,12 +126,12 @@ public:
     InterpolationPolicy _method;
     float               _extrapolation_value;
 
-    std::vector<std::unique_ptr<CLScale>>      _scale;
-    std::vector<std::unique_ptr<CLCopyKernel>> _copy;
-    std::vector<std::unique_ptr<CLTensor>>     _crop_results;
-    std::vector<std::unique_ptr<CLTensor>>     _scaled_results;
+    std::vector<std::unique_ptr<CLScale>>  _scale;
+    std::vector<std::unique_ptr<CLCopy>>   _copy;
+    std::vector<std::unique_ptr<CLTensor>> _crop_results;
+    std::vector<std::unique_ptr<CLTensor>> _scaled_results;
 
-    std::vector<std::unique_ptr<ICLKernel>> _internal_kernels;
+    std::vector<std::unique_ptr<IFunction>> _internal_functions;
 };
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_CL_CROP_RESIZE_H */
diff --git a/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
index 3ebc858d32..6c1302fbf7 100644
--- a/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
+++ b/arm_compute/runtime/CL/functions/CLDeconvolutionLayerUpsample.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,7 +25,7 @@
 #define ARM_COMPUTE_CLDECONVOLUTIONLAYERUPSAMPLE_H
 
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <memory>
@@ -35,13 +35,12 @@ 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:
  *
- * -# @ref CLMemsetKernel
+ * -# @ref CLFill
  * -# @ref CLDeconvolutionLayerUpsampleKernel
  */
 class CLDeconvolutionLayerUpsample : public IFunction
@@ -90,7 +89,7 @@ public:
 
 private:
     std::unique_ptr<CLDeconvolutionLayerUpsampleKernel> _upsample;
-    std::unique_ptr<CLMemsetKernel>                     _memset;
+    CLFill                                              _fill;
     ICLTensor                                          *_output;
 };
 } // namespace arm_compute
diff --git a/arm_compute/runtime/CL/functions/CLFill.h b/arm_compute/runtime/CL/functions/CLFill.h
index fef8324432..9a27d158a6 100644
--- a/arm_compute/runtime/CL/functions/CLFill.h
+++ b/arm_compute/runtime/CL/functions/CLFill.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,32 +24,63 @@
 #ifndef ARM_COMPUTE_CLFILL_H
 #define ARM_COMPUTE_CLFILL_H
 
-#include "arm_compute/core/PixelValue.h"
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include "arm_compute/core/Window.h"
+#include "arm_compute/runtime/IFunction.h"
+#include <memory>
 
 namespace arm_compute
 {
 class CLCompileContext;
 class ICLTensor;
 
-/** Function to run @ref CLMemsetKernel to fill a tensor with a scalar value */
-class CLFill : public ICLSimpleFunction
+/** Basic function to run @ref opencl::kernels::ClFillKernel */
+class CLFill : public IFunction
 {
 public:
-    /** Initialize the function
+    /** Constructor */
+    CLFill();
+    /** Destructor */
+    ~CLFill();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLFill(const CLFill &) = delete;
+    /** Default move constructor */
+    CLFill(CLFill &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLFill &operator=(const CLFill &) = delete;
+    /** Default move assignment operator */
+    CLFill &operator=(CLFill &&);
+    /** Initialize the kernel's tensor and filling value
      *
-     * @param[in,out] tensor         Source tensor. Data types supported: All.
-     * @param[in]     constant_value Constant value to use to fill tensor.
+     * @param[in,out] tensor         Input tensor to fill. Supported data types: All.
+     * @param[in]     constant_value The value used to fill the planes of the tensor
+     * @param[in]     window         Window to be used in case setting only part of a tensor. Default is nullptr.
      */
-    void configure(ICLTensor *tensor, PixelValue constant_value);
-    /** Initialize the function
+    void configure(ICLTensor *tensor, const PixelValue &constant_value, Window *window = nullptr);
+    /** Initialise the kernel's tensor and filling value
      *
      * @param[in]     compile_context The compile context to be used.
-     * @param[in,out] tensor          Source tensor. Data types supported: All.
-     * @param[in]     constant_value  Constant value to use to fill tensor.
+     * @param[in,out] tensor          Input tensor to fill. Supported data types: All.
+     * @param[in]     constant_value  The value used to fill the planes of the tensor
+     * @param[in]     window          Window to be used in case setting only part of a tensor. Default is nullptr.
      */
-    void configure(const CLCompileContext &compile_context, ICLTensor *tensor, PixelValue constant_value);
+    void configure(const CLCompileContext &compile_context, ICLTensor *tensor, const PixelValue &constant_value, Window *window = nullptr);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLFill
+     *
+     * @param[in] tensor         Source tensor info. Data types supported: All.
+     * @param[in] constant_value The value used to fill the planes of the tensor
+     * @param[in] window         Window to be used in case setting only part of a tensor. Default is nullptr.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLFILL_H */
diff --git a/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h b/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
index cf5fd500a0..4d6bc66487 100644
--- a/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGenerateProposalsLayer.h
@@ -27,6 +27,7 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLPermute.h"
 #include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
 #include "arm_compute/runtime/CPP/CPPScheduler.h"
 #include "arm_compute/runtime/CPP/functions/CPPBoxWithNonMaximaSuppressionLimit.h"
@@ -42,7 +43,6 @@ class CLBoundingBoxTransformKernel;
 class CLDequantizationLayerKernel;
 class CLComputeAllAnchorsKernel;
 class CLPadLayerKernel;
-class CLPermuteKernel;
 class CLQuantizationLayerKernel;
 class ICLTensor;
 class ITensorInfo;
@@ -137,9 +137,9 @@ private:
     MemoryGroup _memory_group;
 
     // OpenCL kernels
-    std::unique_ptr<CLPermuteKernel>              _permute_deltas_kernel;
+    CLPermute                                     _permute_deltas;
     CLReshapeLayer                                _flatten_deltas;
-    std::unique_ptr<CLPermuteKernel>              _permute_scores_kernel;
+    CLPermute                                     _permute_scores;
     CLReshapeLayer                                _flatten_scores;
     std::unique_ptr<CLComputeAllAnchorsKernel>    _compute_anchors_kernel;
     std::unique_ptr<CLBoundingBoxTransformKernel> _bounding_box_kernel;
diff --git a/arm_compute/runtime/CL/functions/CLLSTMLayer.h b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
index 017f26aa1e..20b068316c 100644
--- a/arm_compute/runtime/CL/functions/CLLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLLSTMLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,7 +30,9 @@
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
 #include "arm_compute/runtime/CL/functions/CLConcatenateLayer.h"
+#include "arm_compute/runtime/CL/functions/CLCopy.h"
 #include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/CL/functions/CLFullyConnectedLayer.h"
 #include "arm_compute/runtime/CL/functions/CLGEMM.h"
 #include "arm_compute/runtime/CL/functions/CLMeanStdDevNormalizationLayer.h"
@@ -44,8 +46,6 @@
 namespace arm_compute
 {
 class CLCompileContext;
-class CLCopyKernel;
-class CLMemsetKernel;
 class CLTransposeKernel;
 class ICLTensor;
 
@@ -239,14 +239,14 @@ private:
     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;
+    CLCopy                             _copy_cell_state;
+    CLCopy                             _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;
+    CLFill                             _ones_fill;
     CLMeanStdDevNormalizationLayer     _mean_std_norm_input_gate;
     CLPixelWiseMultiplication          _pixelwise_mul_input_gate_coeff;
     CLArithmeticAddition               _accum_input_gate_bias;
diff --git a/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h b/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
index 693862fb89..24d620d372 100644
--- a/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
+++ b/arm_compute/runtime/CL/functions/CLMaxUnpoolingLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,6 +25,7 @@
 #define ARM_COMPUTE_CLMAXUNPOOLINGLAYER_H
 
 #include "arm_compute/core/Error.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <memory>
@@ -35,12 +36,11 @@ class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 class CLMaxUnpoolingLayerKernel;
-class CLMemsetKernel;
 struct PoolingLayerInfo;
 
 /** Function to perform MaxUnpooling. This function calls the following OpenCL kernels:
  *
- * -# @ref CLMemsetKernel
+ * -# @ref CLFill
  * -# @ref CLMaxUnpoolingLayerKernel
  */
 class CLMaxUnpoolingLayer : public IFunction
@@ -99,7 +99,7 @@ public:
     void run() override;
 
 private:
-    std::unique_ptr<CLMemsetKernel>            _memset_kernel;
+    CLFill                                     _fill;
     std::unique_ptr<CLMaxUnpoolingLayerKernel> _unpooling_layer_kernel;
 };
 }
diff --git a/arm_compute/runtime/CL/functions/CLPadLayer.h b/arm_compute/runtime/CL/functions/CLPadLayer.h
index 2bbde30fc2..dae95f63e6 100644
--- a/arm_compute/runtime/CL/functions/CLPadLayer.h
+++ b/arm_compute/runtime/CL/functions/CLPadLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,19 +26,20 @@
 
 #include "arm_compute/core/Error.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLCopy.h"
+#include "arm_compute/runtime/CL/functions/CLPermute.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:
  *
  *  -# @ref CLPadLayerKernel if there is padding to be added
- *  -# @ref CLCopyKernel otherwise
+ *  -# @ref CLCopy otherwise
  */
 class CLPadLayer : public IFunction
 {
@@ -100,7 +101,7 @@ private:
     void configure_reflect_mode(ICLTensor *input, ICLTensor *output);
 
     std::unique_ptr<CLPadLayerKernel> _pad_kernel;
-    std::unique_ptr<CLCopyKernel>     _copy_kernel;
+    CLCopy                            _copy;
     bool                              _perform_pad;
 };
 } // namespace arm_compute
diff --git a/arm_compute/runtime/CL/functions/CLPermute.h b/arm_compute/runtime/CL/functions/CLPermute.h
index 50e81da7c4..bcd9566fbf 100644
--- a/arm_compute/runtime/CL/functions/CLPermute.h
+++ b/arm_compute/runtime/CL/functions/CLPermute.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,9 +25,9 @@
 #define ARM_COMPUTE_CLPERMUTE_H
 
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include "arm_compute/runtime/IFunction.h"
 
-#include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
@@ -35,10 +35,22 @@ class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 
-/** Basic function to execute an @ref CLPermuteKernel. */
-class CLPermute : public ICLSimpleFunction
+/** Basic function to execute an @ref opencl::kernels::ClPermuteKernel. */
+class CLPermute : public IFunction
 {
 public:
+    /** Constructor */
+    CLPermute();
+    /** Destructor */
+    ~CLPermute();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLPermute(const CLPermute &) = delete;
+    /** Default move constructor */
+    CLPermute(CLPermute &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLPermute &operator=(const CLPermute &) = delete;
+    /** Default move assignment operator */
+    CLPermute &operator=(CLPermute &&);
     /** Set the input and output tensors.
      *
      * @note Arbitrary permutation vectors are supported with rank not greater than 4
@@ -69,6 +81,13 @@ public:
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm);
+
+    // Inherited methods overridden:
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLPERMUTE_H */
diff --git a/arm_compute/runtime/CL/functions/CLQLSTMLayer.h b/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
index a8f9221b3d..954f224424 100644
--- a/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
+++ b/arm_compute/runtime/CL/functions/CLQLSTMLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
+#include "arm_compute/runtime/CL/functions/CLCopy.h"
 #include "arm_compute/runtime/CL/functions/CLElementwiseOperations.h"
 #include "arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h"
 #include "arm_compute/runtime/CL/functions/CLGEMMLowpOutputStage.h"
@@ -38,7 +39,6 @@ namespace arm_compute
 {
 // Forward declarations
 class CLCompileContext;
-class CLCopyKernel;
 class ICLTensor;
 class CLGEMMLowpMatrixAReductionKernel;
 class CLQLSTMLayerNormalizationKernel;
@@ -49,12 +49,12 @@ class ITensorInfo;
  * This function calls the following CL functions/kernels:
  *
  * -# @ref CLActivationLayer                                     Activation functions (tanh and logistic)
- * -# @ref CLCopyKernel                                          Copy kernel for copying output_state_out to output
- * -# @ref CLArithmeticAddition                  Elementwise addition and subtraction
+ * -# @ref CLCopy                                                Copy function for copying output_state_out to output
+ * -# @ref CLArithmeticAddition                                  Elementwise addition and subtraction
  * -# @ref CLGEMMLowpMatrixMultiplyCore                          Quantized matrix multiplication core. Accumulators are 32-bit integers
  * -# @ref CLGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint   Convert 32-bit integers into QSYMM16
  * -# @ref CLGEMMLowpMatrixAReductionKernel                      For precomputing effective biases to use
- * -# @ref CLPixelWiseMultiplication                       Elementwise multiplication
+ * -# @ref CLPixelWiseMultiplication                             Elementwise multiplication
  * -# @ref CLTranspose                                           Transpose function for reshaping the weights
  * */
 class CLQLSTMLayer : public IFunction
@@ -354,7 +354,7 @@ private:
     CLArithmeticAddition                              _accumulate_projection{};
     CLActivationLayer                                 _projection_clip{};
     std::array<std::unique_ptr<CLQLSTMLayerNormalizationKernel>, _layer_norm_count> _layer_norms;
-    std::unique_ptr<CLCopyKernel> _copy_output;
+    CLCopy _copy_output;
 
     TensorCopyKernel _projection_bias_copy{};
     TensorCopyKernel _projection_output_to_accumulate_copy{};
diff --git a/arm_compute/runtime/CL/functions/CLRNNLayer.h b/arm_compute/runtime/CL/functions/CLRNNLayer.h
index ff3fb5449b..50575daaa3 100644
--- a/arm_compute/runtime/CL/functions/CLRNNLayer.h
+++ b/arm_compute/runtime/CL/functions/CLRNNLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
 #include "arm_compute/runtime/CL/functions/CLActivationLayer.h"
+#include "arm_compute/runtime/CL/functions/CLCopy.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"
@@ -34,7 +35,6 @@
 
 namespace arm_compute
 {
-class CLCopyKernel;
 class ICLTensor;
 
 /** Basic function to run @ref CLRNNLayer */
@@ -93,16 +93,16 @@ public:
     void prepare() override;
 
 private:
-    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;
+    MemoryGroup           _memory_group;
+    CLGEMM                _gemm_state_f;
+    CLArithmeticAddition  _add_kernel;
+    CLActivationLayer     _activation;
+    CLFullyConnectedLayer _fully_connected_kernel;
+    CLCopy                _copy;
+    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/CLReshapeLayer.h b/arm_compute/runtime/CL/functions/CLReshapeLayer.h
index b4d52ec8cf..60ed81680e 100644
--- a/arm_compute/runtime/CL/functions/CLReshapeLayer.h
+++ b/arm_compute/runtime/CL/functions/CLReshapeLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/runtime/CL/ICLOperator.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include <memory>
 
 namespace arm_compute
 {
@@ -33,7 +34,7 @@ class CLCompileContext;
 class ICLTensor;
 class ITensorInfo;
 
-/** Basic function to run @ref CLReshapeLayerKernel */
+/** Basic function to run @ref opencl::kernels::ClReshapeKernel */
 class CLReshapeLayer : public IFunction
 {
 public:
@@ -79,30 +80,5 @@ private:
     struct Impl;
     std::unique_ptr<Impl> _impl;
 };
-
-namespace experimental
-{
-/** Basic function to run @ref CLReshapeLayerKernel */
-class CLReshape : public ICLOperator
-{
-public:
-    /** Initialise the kernel's inputs and outputs
-     *
-     * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           Input tensor info. Data type supported: All
-     * @param[out] output          Output info. Data type supported: Same as @p input
-     */
-    void configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output);
-
-    /** Static function to check if given info will lead to a valid configuration of @ref CLReshapeLayer
-     *
-     * @param[in] input  Input tensor info. Data type supported: All
-     * @param[in] output Output tensor info. Data type supported: Same as @p input
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output);
-};
-} // namespace experimental
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLRESHAPELAYER_H */
diff --git a/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
index 5c5e5bed9a..dc02fa1363 100644
--- a/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
+++ b/arm_compute/runtime/CL/functions/CLSpaceToBatchLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -26,6 +26,7 @@
 
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/IFunction.h"
 
 #include <memory>
@@ -33,14 +34,13 @@
 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:
  *
- *  -# @ref CLMemsetKernel
+ *  -# @ref CLFill
  *  -# @ref CLSpaceToBatchLayerKernel
  */
 class CLSpaceToBatchLayer : public IFunction
@@ -125,7 +125,7 @@ public:
 
 private:
     std::unique_ptr<CLSpaceToBatchLayerKernel> _space_to_batch_kernel; /**< SpaceToBatch kernel to run */
-    std::unique_ptr<CLMemsetKernel>            _memset_kernel;         /**< Memset kernel to run */
+    CLFill                                     _fill;                  /**< Fill function to run */
     bool                                       _has_padding;           /**< Flag to check if the output has padding */
 };
 } // namespace arm_compute
diff --git a/docs/00_introduction.dox b/docs/00_introduction.dox
index ab2495dbff..2f239960c3 100644
--- a/docs/00_introduction.dox
+++ b/docs/00_introduction.dox
@@ -191,7 +191,7 @@ v20.11 Public major release
    - @ref CLFuseBatchNormalizationKernel
    - @ref CLDepthwiseConvolutionLayerNativeKernel
    - @ref CLDepthConvertLayerKernel
-   - @ref CLCopyKernel
+   - CLCopyKernel
    - @ref CLDepthwiseConvolutionLayer3x3NHWCKernel
    - CLActivationLayerKernel
    - @ref CLWinogradFilterTransformKernel
@@ -434,7 +434,7 @@ v20.08 Public major release
    - @ref CLArgMinMaxLayerKernel
  - Added new data type U8 support for:
    - @ref NECropKernel
-   - @ref CLCropKernel
+   - CLCropKernel
  - Added aligh_corner support for nearest neighbor interpolation in:
    - @ref NEScaleKernel
    - @ref CLScaleKernel
@@ -777,7 +777,7 @@ v19.05 Public major release
     - @ref NEFFTConvolutionLayer
  - New OpenCL kernels / functions:
     - @ref CLComplexPixelWiseMultiplicationKernel / @ref CLComplexPixelWiseMultiplication
-    - @ref CLCropKernel / @ref CLCropResize
+    - CLCropKernel / @ref CLCropResize
     - @ref CLDeconvolutionReshapeOutputKernel
     - @ref CLFFTDigitReverseKernel
     - @ref CLFFTRadixStageKernel
@@ -1019,7 +1019,7 @@ v18.05 Public major release
  - New OpenCL kernels / functions:
     - @ref CLChannelShuffleLayer / @ref CLChannelShuffleLayerKernel
     - @ref CLConvertFullyConnectedWeightsKernel / @ref CLConvertFullyConnectedWeights
-    - @ref CLCopy / @ref CLCopyKernel
+    - @ref CLCopy / CLCopyKernel
     - @ref CLLSTMLayer
     - @ref CLRNNLayer
     - CLWidthConcatenateLayer / CLWidthConcatenateLayerKernel
@@ -1103,7 +1103,7 @@ v18.01 Public maintenance release
  - Various bug fixes
  - Added some of the missing validate() methods
  - Added @ref CLDeconvolutionLayerUpsampleKernel / @ref CLDeconvolutionLayer @ref CLDeconvolutionLayerUpsample
- - Added @ref CLPermuteKernel / @ref CLPermute
+ - Added CLPermuteKernel / @ref CLPermute
  - Added method to clean the programs cache in the CL Kernel library.
  - Added @ref GCArithmeticAdditionKernel / @ref GCArithmeticAddition
  - Added @ref GCDepthwiseConvolutionLayer3x3Kernel / @ref GCDepthwiseConvolutionLayer3x3
@@ -1218,7 +1218,7 @@ v17.09 Public major release
     - @ref CLQuantizationLayerKernel @ref CLMinMaxLayerKernel / @ref CLQuantizationLayer
     - @ref CLROIPoolingLayerKernel / @ref CLROIPoolingLayer
     - @ref CLReductionOperationKernel / @ref CLReductionOperation
-    - @ref CLReshapeLayerKernel / @ref CLReshapeLayer
+    - CLReshapeLayerKernel / @ref CLReshapeLayer
 
 v17.06 Public major release
  - Various bug fixes
diff --git a/docs/04_adding_operator.dox b/docs/04_adding_operator.dox
index 9e6f3751b8..f311fb4d51 100644
--- a/docs/04_adding_operator.dox
+++ b/docs/04_adding_operator.dox
@@ -117,7 +117,7 @@ Each kernel will have to implement the method:
 
 The structure of the kernel .cpp file should be similar to the next ones.
 For OpenCL:
-@snippet src/core/CL/kernels/CLReshapeLayerKernel.cpp CLReshapeLayerKernel Kernel
+@snippet src/core/gpu/cl/kernels/ClReshapeKernel.cpp ClReshapeKernel Kernel
 The run will call the function defined in the .cl file.
 
 For the NEON backend case:
diff --git a/src/core/CL/CLKernels.h b/src/core/CL/CLKernels.h
index ac051684a1..7383dce40f 100644
--- a/src/core/CL/CLKernels.h
+++ b/src/core/CL/CLKernels.h
@@ -42,8 +42,6 @@
 #include "src/core/CL/kernels/CLComparisonKernel.h"
 #include "src/core/CL/kernels/CLConvertFullyConnectedWeightsKernel.h"
 #include "src/core/CL/kernels/CLConvolutionKernel.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
-#include "src/core/CL/kernels/CLCropKernel.h"
 #include "src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
 #include "src/core/CL/kernels/CLDeconvolutionReshapeOutputKernel.h"
 #include "src/core/CL/kernels/CLDepthConvertLayerKernel.h"
@@ -97,7 +95,6 @@
 #include "src/core/CL/kernels/CLMeanStdDevKernel.h"
 #include "src/core/CL/kernels/CLMeanStdDevNormalizationKernel.h"
 #include "src/core/CL/kernels/CLMedian3x3Kernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 #include "src/core/CL/kernels/CLMinMaxLayerKernel.h"
 #include "src/core/CL/kernels/CLMinMaxLocationKernel.h"
 #include "src/core/CL/kernels/CLNonLinearFilterKernel.h"
@@ -105,7 +102,6 @@
 #include "src/core/CL/kernels/CLNormalizationLayerKernel.h"
 #include "src/core/CL/kernels/CLNormalizePlanarYUVLayerKernel.h"
 #include "src/core/CL/kernels/CLPadLayerKernel.h"
-#include "src/core/CL/kernels/CLPermuteKernel.h"
 #include "src/core/CL/kernels/CLPixelWiseMultiplicationKernel.h"
 #include "src/core/CL/kernels/CLPriorBoxLayerKernel.h"
 #include "src/core/CL/kernels/CLQLSTMLayerNormalizationKernel.h"
@@ -116,7 +112,6 @@
 #include "src/core/CL/kernels/CLReductionOperationKernel.h"
 #include "src/core/CL/kernels/CLRemapKernel.h"
 #include "src/core/CL/kernels/CLReorgLayerKernel.h"
-#include "src/core/CL/kernels/CLReshapeLayerKernel.h"
 #include "src/core/CL/kernels/CLReverseKernel.h"
 #include "src/core/CL/kernels/CLScaleKernel.h"
 #include "src/core/CL/kernels/CLScharr3x3Kernel.h"
diff --git a/src/core/CL/kernels/CLCopyKernel.h b/src/core/CL/kernels/CLCopyKernel.h
deleted file mode 100644
index 9a20b88884..0000000000
--- a/src/core/CL/kernels/CLCopyKernel.h
+++ /dev/null
@@ -1,83 +0,0 @@
-/*
- * Copyright (c) 2018-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLCOPYKERNEL_H
-#define ARM_COMPUTE_CLCOPYKERNEL_H
-
-#include "arm_compute/core/Types.h"
-#include "src/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** OpenCL kernel to perform a copy between two tensors */
-class CLCopyKernel : public ICLKernel
-{
-public:
-    /** Default constructor */
-    CLCopyKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers). */
-    CLCopyKernel(const CLCopyKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers). */
-    CLCopyKernel &operator=(const CLCopyKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLCopyKernel(CLCopyKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLCopyKernel &operator=(CLCopyKernel &&) = default;
-    /** Initialize the kernel's input, output.
-     *
-     * @param[in]  input         Source tensor. Data types supported: All.
-     * @param[out] output        Destination tensor. Data types supported: same as @p input.
-     * @param[in]  output_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
-     */
-    void configure(const ICLTensor *input, ICLTensor *output, Window *output_window = nullptr);
-    /** Initialize the kernel's input, output.
-     *
-     * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           Source tensor. Data types supported: All.
-     * @param[out] output          Destination tensor. Data types supported: same as @p input.
-     * @param[in]  output_window   (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
-     */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, Window *output_window = nullptr);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLCopyKernel
-     *
-     * @param[in] input         Source tensor info. Data types supported: All.
-     * @param[in] output        Destination tensor info. Data types supported: same as @p input.
-     * @param[in] output_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, Window *output_window = nullptr);
-
-    // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
-    const ICLTensor *_input;
-    ICLTensor       *_output;
-    Window           _output_window;
-    bool             _has_output_window;
-};
-} // namespace arm_compute
-#endif /*ARM_COMPUTE_CLCOPYKERNEL_H */
diff --git a/src/core/CL/kernels/CLPermuteKernel.h b/src/core/CL/kernels/CLPermuteKernel.h
deleted file mode 100644
index d1bb875d7a..0000000000
--- a/src/core/CL/kernels/CLPermuteKernel.h
+++ /dev/null
@@ -1,90 +0,0 @@
-/*
- * Copyright (c) 2018-2020 Arm Limited.
- *
- * SPDX-License-Identifier: MIT
- *
- * Permission is hereby granted, free of charge, to any person obtaining a copy
- * of this software and associated documentation files (the "Software"), to
- * deal in the Software without restriction, including without limitation the
- * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
- * sell copies of the Software, and to permit persons to whom the Software is
- * furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in all
- * copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-#ifndef ARM_COMPUTE_CLPERMUTEKERNEL_H
-#define ARM_COMPUTE_CLPERMUTEKERNEL_H
-
-#include "src/core/CL/ICLKernel.h"
-
-namespace arm_compute
-{
-class ICLTensor;
-
-/** OpenCL kernel to perform tensor permutation.
- *
- * Permutes given a permutation vector
- */
-class CLPermuteKernel : public ICLKernel
-{
-public:
-    /** Default constructor */
-    CLPermuteKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLPermuteKernel(const CLPermuteKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLPermuteKernel &operator=(const CLPermuteKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLPermuteKernel(CLPermuteKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLPermuteKernel &operator=(CLPermuteKernel &&) = default;
-    /** Set the input and output of the kernel.
-     *
-     * @note Arbitrary permutation vectors are supported with rank not greater than 4
-     *
-     * @param[in] input  The input tensor to permute. Data types supported: All.
-     * @param[in] output The output tensor. Data types supported: Same as @p input
-     * @param[in] perm   Permutation vector
-     */
-    void configure(const ICLTensor *input, ICLTensor *output, const PermutationVector &perm);
-    /** Set the input and output of the kernel.
-     *
-     * @note Arbitrary permutation vectors are supported with rank not greater than 4
-     *
-     * @param[in] compile_context The compile context to be used.
-     * @param[in] input           The input tensor to permute. Data types supported: All.
-     * @param[in] output          The output tensor. Data types supported: Same as @p input
-     * @param[in] perm            Permutation vector
-     */
-    void configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PermutationVector &perm);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLPermuteKernel
-     *
-     * @note Arbitrary permutation vectors are supported with rank not greater than 4
-     *
-     * @param[in] input  First tensor input info. Data types supported: All.
-     * @param[in] output Output tensor info. Data types supported: same as @p input.
-     * @param[in] perm   Permutation vector
-     *
-     * @return a status
-     */
-    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm);
-
-    // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
-
-private:
-    const ICLTensor *_input;
-    ICLTensor        *_output;
-    PermutationVector _perm;
-};
-} // arm_compute
-#endif /*ARM_COMPUTE_CLPERMUTEKERNEL_H */
diff --git a/src/core/CL/kernels/CLCopyKernel.cpp b/src/core/gpu/cl/kernels/ClCopyKernel.cpp
index ca38b65df4..d6c87f8fad 100644
--- a/src/core/CL/kernels/CLCopyKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClCopyKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,36 +21,45 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLCopyKernel.h"
+#include "src/core/gpu/cl/kernels/ClCopyKernel.h"
 
 #include "arm_compute/core/CL/CLHelpers.h"
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
 #include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/CLValidate.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
 #include "support/StringSupport.h"
 
 namespace arm_compute
 {
+namespace opencl
+{
+namespace kernels
+{
 namespace
 {
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, Window *output_window = nullptr)
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, Window *dst_window = nullptr)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
 
-    // Validate output if initialized
-    if(output->total_size() != 0)
+    // Validate dst if initialized
+    if(dst->total_size() != 0)
     {
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-        if(output_window == nullptr)
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+        if(dst_window == nullptr)
         {
-            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(input->tensor_shape(), output->tensor_shape());
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(src->tensor_shape(), dst->tensor_shape());
         }
         else
         {
-            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(input->tensor_shape(), output_window->shape());
+            ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(src->tensor_shape(), dst_window->shape());
         }
     }
 
@@ -59,56 +68,43 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, W
 
 } // namespace
 
-CLCopyKernel::CLCopyKernel()
-    : _input(nullptr), _output(nullptr), _output_window(), _has_output_window(false)
-{
-}
-
-void CLCopyKernel::configure(const ICLTensor *input, ICLTensor *output, Window *output_window)
+void ClCopyKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Window *dst_window)
 {
-    configure(CLKernelLibrary::get().get_compile_context(), input, output, output_window);
-}
-
-void CLCopyKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, Window *output_window)
-{
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), output_window));
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, dst_window));
 
-    auto padding_info = get_padding_info({ input, output });
-
-    _input  = input;
-    _output = output;
+    auto padding_info = get_padding_info({ src, dst });
 
     // Create kernel
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));
 
     // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*(output->info()), *(input->info()));
+    auto_init_if_empty(*dst, *src);
 
     // Configure window
-    const unsigned int vec_size_x = adjust_vec_size(16 / input->info()->element_size(), input->info()->dimension(0));
+    const unsigned int vec_size_x = adjust_vec_size(16 / src->element_size(), src->dimension(0));
 
-    const Window win_config = calculate_max_window(*(input->info()), Steps(vec_size_x));
+    const Window win_config = calculate_max_window(*src, Steps(vec_size_x));
 
-    if(output_window != nullptr)
+    if(dst_window != nullptr)
     {
-        _has_output_window             = true;
-        _output_window                 = Window(*output_window);
-        const int  width_x             = output_window->num_iterations(0);
+        _has_dst_window                = true;
+        _dst_window                    = Window(*dst_window);
+        const int  width_x             = dst_window->num_iterations(0);
         const int  vec_size_x_leftover = width_x % vec_size_x;
         const bool multi_access_x      = width_x >= static_cast<int32_t>(vec_size_x);
 
         if(multi_access_x)
         {
-            _output_window.set(Window::DimX, Window::Dimension(output_window->x().start(), ceil_to_multiple(output_window->x().end(), vec_size_x), vec_size_x));
+            _dst_window.set(Window::DimX, Window::Dimension(dst_window->x().start(), ceil_to_multiple(dst_window->x().end(), vec_size_x), vec_size_x));
         }
 
         build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
     }
     else
     {
-        const int width_x             = input->info()->tensor_shape().x();
+        const int width_x             = src->tensor_shape().x();
         const int vec_size_x_leftover = width_x % vec_size_x;
 
         build_opts.add_option("-DVEC_SIZE_LEFTOVER=" + support::cpp11::to_string(vec_size_x_leftover));
@@ -125,32 +121,35 @@ void CLCopyKernel::configure(const CLCompileContext &compile_context, const ICLT
     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
 }
 
-Status CLCopyKernel::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output, Window *output_window)
+Status ClCopyKernel::validate(const arm_compute::ITensorInfo *src, const arm_compute::ITensorInfo *dst, Window *dst_window)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, output_window));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, dst_window));
 
     return Status{};
 }
 
-void CLCopyKernel::run(const Window &window, cl::CommandQueue &queue)
+void ClCopyKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
     Window slice;
 
-    if(_has_output_window)
+    if(_has_dst_window)
     {
         slice            = window.first_slice_window_3D();
-        Window out_slice = _output_window.first_slice_window_3D();
+        Window out_slice = _dst_window.first_slice_window_3D();
         do
         {
             unsigned int idx = 0;
-            add_3D_tensor_argument(idx, _input, slice);
-            add_3D_tensor_argument(idx, _output, out_slice);
+            add_3D_tensor_argument(idx, src, slice);
+            add_3D_tensor_argument(idx, dst, out_slice);
             enqueue(queue, *this, slice, lws_hint());
         }
-        while(window.slide_window_slice_3D(slice) && _output_window.slide_window_slice_3D(out_slice));
+        while(window.slide_window_slice_3D(slice) && _dst_window.slide_window_slice_3D(out_slice));
     }
     else
     {
@@ -159,11 +158,13 @@ void CLCopyKernel::run(const Window &window, cl::CommandQueue &queue)
         do
         {
             unsigned int idx = 0;
-            add_3D_tensor_argument(idx, _input, slice);
-            add_3D_tensor_argument(idx, _output, slice);
+            add_3D_tensor_argument(idx, src, slice);
+            add_3D_tensor_argument(idx, dst, slice);
             enqueue(queue, *this, slice, lws_hint());
         }
         while(collapsed.slide_window_slice_3D(slice));
     }
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClCopyKernel.h b/src/core/gpu/cl/kernels/ClCopyKernel.h
new file mode 100644
index 0000000000..9d5457d72d
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClCopyKernel.h
@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2018-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_COPY_KERNEL_H
+#define ARM_COMPUTE_CL_COPY_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** OpenCL kernel to perform a copy between two tensors */
+class ClCopyKernel : public ICLKernel
+{
+public:
+    ClCopyKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCopyKernel);
+    /** Initialize the kernel's src, dst.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info. Data types supported: All.
+     * @param[out] dst             Destination tensor info. Data types supported: same as @p src.
+     * @param[in]  dst_window      (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Window *dst_window = nullptr);
+    /** Static function to check if given info will lead to a valid configuration of @ref ClCopyKernel
+     *
+     * @param[in] src        Source tensor info. Data types supported: All.
+     * @param[in] dst        Destination tensor info. Data types supported: same as @p src.
+     * @param[in] dst_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, Window *dst_window = nullptr);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    Window _dst_window{};
+    bool   _has_dst_window{};
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CL_COPY_KERNEL_H */
diff --git a/src/core/CL/kernels/CLCropKernel.cpp b/src/core/gpu/cl/kernels/ClCropKernel.cpp
index 9cf15ff93b..20e94b2876 100644
--- a/src/core/CL/kernels/CLCropKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClCropKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,57 +21,55 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLCropKernel.h"
+#include "src/core/gpu/cl/kernels/ClCropKernel.h"
 
 #include "arm_compute/core/CL/CLHelpers.h"
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/IAccessWindow.h"
 #include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
 #include "src/core/CPP/Validate.h"
 #include "src/core/helpers/WindowHelpers.h"
-
+#include "support/Cast.h"
 #include "support/StringSupport.h"
 
 #include <map>
 
 namespace arm_compute
 {
-CLCropKernel::CLCropKernel()
-    : _input(nullptr), _output(nullptr), _start(), _batch_index(0), _extrapolation_value(0)
+namespace opencl
 {
-}
-
-void CLCropKernel::configure(const ICLTensor *input, ICLTensor *output, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *output_window)
+namespace kernels
 {
-    configure(CLKernelLibrary::get().get_compile_context(), input, output, start, end, batch_index, extrapolation_value, output_window);
+void ClCropKernel::configure(const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *dst_window)
+{
+    configure(CLKernelLibrary::get().get_compile_context(), src, dst, start, end, batch_index, extrapolation_value, dst_window);
 }
 
-void CLCropKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, Coordinates2D start, Coordinates2D end, uint32_t batch_index,
-                             float extrapolation_value, Window *output_window)
+void ClCropKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index,
+                             float extrapolation_value, Window *dst_window)
 {
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate(input->info(), output->info(), start, end, batch_index, extrapolation_value, output_window));
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(validate(src, dst, start, end, batch_index, extrapolation_value, dst_window));
 
-    _input               = input;
-    _output              = output;
     _start               = start;
     _batch_index         = batch_index;
     _extrapolation_value = extrapolation_value;
 
     const int vec_size_x = 4;
     // Create and update the window (if needed)
-    Window win = calculate_max_window(*output->info());
+    Window win = calculate_max_window(*dst);
 
-    if(output_window != nullptr)
+    if(dst_window != nullptr)
     {
-        ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(win, *output_window);
-        win = *output_window;
+        ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(win, *dst_window);
+        win = *dst_window;
     }
 
-    const int  output_width_x = win.num_iterations(0);
-    const bool multi_access_x = output_width_x >= vec_size_x;
-    const bool remainder_x    = output_width_x % vec_size_x > 0;
+    const int  dst_width_x    = win.num_iterations(0);
+    const bool multi_access_x = dst_width_x >= vec_size_x;
+    const bool remainder_x    = dst_width_x % vec_size_x > 0;
 
     if(multi_access_x)
     {
@@ -82,53 +80,58 @@ void CLCropKernel::configure(const CLCompileContext &compile_context, const ICLT
 
     // Create kernel
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(input->info()->data_type()));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_type_from_data_type(src->data_type()));
     build_opts.add_option_if(multi_access_x, "-DVEC_SIZE=" + support::cpp11::to_string(vec_size_x));
-    build_opts.add_option_if(multi_access_x && remainder_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(output_width_x - vec_size_x, 0)));
+    build_opts.add_option_if(multi_access_x && remainder_x, "-DLAST_ACCESSED_X=" + support::cpp11::to_string(std::max<int>(dst_width_x - vec_size_x, 0)));
     build_opts.add_option_if(start.x > end.x, "-DWIDTH_FLIPPED=");
     build_opts.add_option_if(start.y > end.y, "-DHEIGHT_FLIPPED=");
     _kernel = create_kernel(compile_context, "crop_tensor", build_opts.options());
 }
 
-Status CLCropKernel::validate(const ITensorInfo *input, const ITensorInfo *output, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *output_window)
+Status ClCropKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *dst_window)
 {
-    ARM_COMPUTE_UNUSED(extrapolation_value, output_window);
-    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(input, DataLayout::NHWC);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().num_dimensions() > 4);
+    ARM_COMPUTE_UNUSED(extrapolation_value, dst_window);
+    ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_LAYOUT_NOT_IN(src, DataLayout::NHWC);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().num_dimensions() > 4);
     ARM_COMPUTE_RETURN_ERROR_ON(start.x < 0 || start.y < 0 || end.x < 0 || end.y < 0);
-    ARM_COMPUTE_RETURN_ERROR_ON(start.x >= static_cast<int32_t>(input->dimension(1)) || start.y >= static_cast<int32_t>(input->dimension(2))
-                                || end.x >= static_cast<int32_t>(input->dimension(1)) || end.y >= static_cast<int32_t>(input->dimension(2)));
-    ARM_COMPUTE_RETURN_ERROR_ON(batch_index >= input->dimension(3));
-    if(output_window != nullptr)
+    ARM_COMPUTE_RETURN_ERROR_ON(start.x >= static_cast<int32_t>(src->dimension(1)) || start.y >= static_cast<int32_t>(src->dimension(2))
+                                || end.x >= static_cast<int32_t>(src->dimension(1)) || end.y >= static_cast<int32_t>(src->dimension(2)));
+    ARM_COMPUTE_RETURN_ERROR_ON(batch_index >= src->dimension(3));
+    if(dst_window != nullptr)
     {
-        ARM_COMPUTE_RETURN_ERROR_ON(output_window->x().step() != 1);
+        ARM_COMPUTE_RETURN_ERROR_ON(dst_window->x().step() != 1);
     }
-    if(output->total_size() > 0)
+    if(dst->total_size() > 0)
     {
-        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(output, DataType::F32);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON(output->num_dimensions() > 3);
+        ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(dst, DataType::F32);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_LAYOUT(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON(dst->num_dimensions() > 3);
     }
     return Status{};
 }
 
-void CLCropKernel::run(const Window &window, cl::CommandQueue &queue)
+void ClCropKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
     Window in_slice = Window();
-    in_slice.use_tensor_dimensions(_input->info()->tensor_shape());
+    in_slice.use_tensor_dimensions(src->info()->tensor_shape());
     in_slice.set(Window::DimX, Window::Dimension(in_slice.x().start(), ceil_to_multiple(in_slice.x().end(), window.x().step()), window.x().step()));
     in_slice.set(3, Window::Dimension(_batch_index, _batch_index + 1, 1));
 
     unsigned int idx = 0;
-    add_3D_tensor_argument(idx, _input, in_slice);
-    add_3D_tensor_argument(idx, _output, window);
+    add_3D_tensor_argument(idx, src, in_slice);
+    add_3D_tensor_argument(idx, dst, window);
     add_argument(idx, _start.x);
     add_argument(idx, _start.y);
     enqueue(queue, *this, window, lws_hint());
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
diff --git a/src/core/gpu/cl/kernels/ClCropKernel.h b/src/core/gpu/cl/kernels/ClCropKernel.h
new file mode 100644
index 0000000000..92f94d471b
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClCropKernel.h
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2019-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CLCROPKERNEL_H
+#define ARM_COMPUTE_CLCROPKERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** OpenCL kernel to perform a copy between two tensors */
+class ClCropKernel : public ICLKernel
+{
+public:
+    ClCropKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClCropKernel);
+    /** Configure kernel
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in]  src                 Source tensor info. Data type supported: All. Data layouts supported: NHWC.
+     * @param[out] dst                 Destination tensor info. Data type supported: F32
+     * @param[in]  start               Coordinates of where to start cropping the image.
+     * @param[in]  end                 Coordinates of where to end cropping the image.
+     * @param[in]  batch_index         Fourth dimension index of the 3D image to crop in @p src.
+     * @param[in]  extrapolation_value Value to be used for values outside of the image. Default is 0.
+     * @param[in]  dst_window          Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
+     */
+    void configure(const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0, Window *dst_window = nullptr);
+    /** Configure kernel
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in]  compile_context     The compile context to be used.
+     * @param[in]  src                 Source tensor info. Data type supported: All. Data layouts supported: NHWC.
+     * @param[out] dst                 Destination tensor info. Data type supported: F32
+     * @param[in]  start               Coordinates of where to start cropping the image.
+     * @param[in]  end                 Coordinates of where to end cropping the image.
+     * @param[in]  batch_index         Fourth dimension index of the 3D image to crop in @p src.
+     * @param[in]  extrapolation_value Value to be used for values outside of the image. Default is 0.
+     * @param[in]  dst_window          Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0,
+                   Window *dst_window = nullptr);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref CLStridedSliceKernel
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in] src                 Source tensor info. Data type supported: All. Data layouts supported: NHWC.
+     * @param[in] dst                 Destination tensor info. Data type supported: F32
+     * @param[in] start               Coordinates of where to start cropping the image.
+     * @param[in] end                 Coordinates of where to end cropping the image.
+     * @param[in] batch_index         Fourth dimension index of the 3D image to crop in @p src.
+     * @param[in] extrapolation_value Value to be used for values outside of the image. Default is 0.
+     * @param[in] dst_window          Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0,
+                           Window *dst_window = nullptr);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    Coordinates2D _start{};
+    uint32_t      _batch_index{};
+    float         _extrapolation_value{};
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CLCROPKERNEL_H */
diff --git a/src/core/CL/kernels/CLMemsetKernel.cpp b/src/core/gpu/cl/kernels/ClFillKernel.cpp
index 2543b07a1a..b194ee549b 100644
--- a/src/core/CL/kernels/CLMemsetKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClFillKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,40 +21,46 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLMemsetKernel.h"
+#include "src/core/gpu/cl/kernels/ClFillKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/CLValidate.h"
+#include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
 #include "support/StringSupport.h"
 
 namespace arm_compute
 {
-CLMemsetKernel::CLMemsetKernel()
-    : ICLKernel(), _tensor(nullptr), _full_window()
+namespace opencl
 {
-}
-
-void CLMemsetKernel::configure(ICLTensor        *tensor,
-                               const PixelValue &constant_value,
-                               Window           *window)
+namespace kernels
+{
+void ClFillKernel::configure(ITensorInfo      *tensor,
+                             const PixelValue &constant_value,
+                             Window           *window)
 {
     configure(CLKernelLibrary::get().get_compile_context(), tensor, constant_value, window);
 }
 
-void CLMemsetKernel::configure(const CLCompileContext &compile_context, ICLTensor *tensor,
-                               const PixelValue &constant_value,
-                               Window           *window)
+void ClFillKernel::configure(const CLCompileContext &compile_context, ITensorInfo *tensor,
+                             const PixelValue &constant_value,
+                             Window           *window)
 {
     ARM_COMPUTE_ERROR_ON_NULLPTR(tensor);
-    ARM_COMPUTE_ERROR_THROW_ON(validate(tensor->info(), constant_value, window));
-
-    _tensor = tensor;
+    ARM_COMPUTE_ERROR_THROW_ON(validate(tensor, constant_value, window));
 
-    const DataType data_type  = tensor->info()->data_type();
-    const int      vec_size_x = 16 / tensor->info()->element_size();
+    const DataType data_type  = tensor->data_type();
+    const int      vec_size_x = 16 / tensor->element_size();
 
     // Create and update the window (if needed)
-    _full_window = calculate_max_window(*tensor->info());
+    _full_window = calculate_max_window(*tensor);
     Window win   = _full_window;
     if(window != nullptr)
     {
@@ -81,7 +87,7 @@ void CLMemsetKernel::configure(const CLCompileContext &compile_context, ICLTenso
     _kernel = create_kernel(compile_context, "memset", build_opts.options());
 }
 
-Status CLMemsetKernel::validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *window)
+Status ClFillKernel::validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *window)
 {
     ARM_COMPUTE_UNUSED(tensor);
     ARM_COMPUTE_UNUSED(constant_value);
@@ -92,11 +98,13 @@ Status CLMemsetKernel::validate(const ITensorInfo *tensor, const PixelValue &con
     return Status{};
 }
 
-void CLMemsetKernel::run(const Window &window, cl::CommandQueue &queue)
+void ClFillKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(ICLKernel::window(), window);
 
+    const auto tensor = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+
     // Collapse all the batches on the third
     Window collapsed = window.collapse_if_possible(_full_window, Window::DimZ);
     Window slice     = collapsed.first_slice_window_3D();
@@ -104,9 +112,11 @@ void CLMemsetKernel::run(const Window &window, cl::CommandQueue &queue)
     do
     {
         unsigned int idx = 0;
-        add_3D_tensor_argument(idx, _tensor, slice);
+        add_3D_tensor_argument(idx, tensor, slice);
         enqueue(queue, *this, slice, lws_hint());
     }
     while(collapsed.slide_window_slice_3D(slice));
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
diff --git a/src/core/CL/kernels/CLMemsetKernel.h b/src/core/gpu/cl/kernels/ClFillKernel.h
index dc103f580f..136c1243fe 100644
--- a/src/core/CL/kernels/CLMemsetKernel.h
+++ b/src/core/gpu/cl/kernels/ClFillKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,50 +21,41 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_CLMEMSETKERNEL_H
-#define ARM_COMPUTE_CLMEMSETKERNEL_H
+#ifndef ARM_COMPUTE_CL_FILL_KERNEL_H
+#define ARM_COMPUTE_CL_FILL_KERNEL_H
 
-#include "arm_compute/core/PixelValue.h"
-#include "arm_compute/core/Types.h"
-#include "src/core/CL/ICLKernel.h"
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
 
 namespace arm_compute
 {
-class ICLTensor;
-
+namespace opencl
+{
+namespace kernels
+{
 /** Interface for filling the planes of a tensor */
-class CLMemsetKernel : public ICLKernel
+class ClFillKernel : public ICLKernel
 {
 public:
-    /** Default constructor */
-    CLMemsetKernel();
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLMemsetKernel(const CLMemsetKernel &) = delete;
-    /** Prevent instances of this class from being copied (As this class contains pointers) */
-    CLMemsetKernel &operator=(const CLMemsetKernel &) = delete;
-    /** Allow instances of this class to be moved */
-    CLMemsetKernel(CLMemsetKernel &&) = default;
-    /** Allow instances of this class to be moved */
-    CLMemsetKernel &operator=(CLMemsetKernel &&) = default;
-    /** Default destructor */
-    ~CLMemsetKernel() = default;
-
+    ClFillKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClFillKernel);
     /** Initialise the kernel's tensor and filling value
      *
-     * @param[in,out] tensor         Input tensor to fill. Supported data types: All.
+     * @param[in,out] tensor         Input tensor info. Supported data types: All.
      * @param[in]     constant_value The value used to fill the planes of the tensor
      * @param[in]     window         Window to be used in case setting only part of a tensor. Default is nullptr.
      */
-    void configure(ICLTensor *tensor, const PixelValue &constant_value, Window *window = nullptr);
+    void configure(ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
     /** Initialise the kernel's tensor and filling value
      *
      * @param[in]     compile_context The compile context to be used.
-     * @param[in,out] tensor          Input tensor to fill. Supported data types: All.
+     * @param[in,out] tensor          Input tensor info. Supported data types: All.
      * @param[in]     constant_value  The value used to fill the planes of the tensor
      * @param[in]     window          Window to be used in case setting only part of a tensor. Default is nullptr.
      */
-    void configure(const CLCompileContext &compile_context, ICLTensor *tensor, const PixelValue &constant_value, Window *window = nullptr);
-    /** Static function to check if given info will lead to a valid configuration of @ref CLMemsetKernel
+    void configure(const CLCompileContext &compile_context, ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
+    /** Static function to check if given info will lead to a valid configuration of @ref ClFillKernel
      *
      * @param[in] tensor         Source tensor info. Data types supported: All.
      * @param[in] constant_value The value used to fill the planes of the tensor
@@ -75,11 +66,12 @@ public:
     static Status validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
 
     // Inherited methods overridden:
-    void run(const Window &window, cl::CommandQueue &queue) override;
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
 
 private:
-    ICLTensor *_tensor;
-    Window     _full_window;
+    Window _full_window{};
 };
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_CLMEMSETRKERNEL_H */
diff --git a/src/core/CL/kernels/CLPermuteKernel.cpp b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
index 07d83bddc2..992c2a89d3 100644
--- a/src/core/CL/kernels/CLPermuteKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,34 +21,43 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLPermuteKernel.h"
+#include "src/core/gpu/cl/kernels/ClPermuteKernel.h"
+
+#include "arm_compute/core/CL/CLHelpers.h"
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Helpers.h"
+#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Utils.h"
+#include "arm_compute/core/Validate.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
+#include "src/core/CL/CLValidate.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
+#include "support/Cast.h"
 #include "support/StringSupport.h"
 
 namespace arm_compute
 {
-CLPermuteKernel::CLPermuteKernel()
-    : _input(nullptr), _output(nullptr), _perm()
+namespace opencl
+{
+namespace kernels
 {
-}
 namespace
 {
-TensorShape get_output_shape(const ITensorInfo *input, const PermutationVector &perm)
+TensorShape get_dst_shape(const ITensorInfo *src, const PermutationVector &perm)
 {
-    TensorShape output_shape = input->tensor_shape();
-    permute(output_shape, perm);
-    return output_shape;
+    TensorShape dst_shape = src->tensor_shape();
+    permute(dst_shape, perm);
+    return dst_shape;
 }
 
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst, const PermutationVector &perm)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->num_dimensions() < 1 || input->num_dimensions() > 4,
-                                    "Permutation upto 4-D input tensor is supported");
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
+    ARM_COMPUTE_RETURN_ERROR_ON_MSG(src->num_dimensions() < 1 || src->num_dimensions() > 4,
+                                    "Permutation upto 4-D src tensor is supported");
     ARM_COMPUTE_RETURN_ERROR_ON_MSG(perm.num_dimensions() < 1 || perm.num_dimensions() > 4,
                                     "Permutation vector size should be less than or equal to 4");
     for(const auto &p : perm)
@@ -56,41 +65,39 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c
         ARM_COMPUTE_RETURN_ERROR_ON_MSG(p >= perm.num_dimensions(), "Permutation vector has invalid values");
     }
 
-    // Validate configured output
-    if(output->total_size() != 0)
+    // Validate configured dst
+    if(dst->total_size() != 0)
     {
-        const TensorShape output_shape = misc::shape_calculator::compute_permutation_output_shape(*input, perm);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(output->tensor_shape(), output_shape);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
+        const TensorShape dst_shape = misc::shape_calculator::compute_permutation_output_shape(*src, perm);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DIMENSIONS(dst->tensor_shape(), dst_shape);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+        ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
     }
     return Status{};
 }
 } // namespace
 
-void CLPermuteKernel::configure(const ICLTensor *input, ICLTensor *output, const PermutationVector &perm)
+void ClPermuteKernel::configure(const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
 {
-    configure(CLKernelLibrary::get().get_compile_context(), input, output, perm);
+    configure(CLKernelLibrary::get().get_compile_context(), src, dst, perm);
 }
 
-void CLPermuteKernel::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PermutationVector &perm)
+void ClPermuteKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
 {
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    auto              padding_info = get_padding_info({ input, output });
-    const TensorShape output_shape = get_output_shape(input->info(), perm);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    auto              padding_info = get_padding_info({ src, dst });
+    const TensorShape dst_shape    = get_dst_shape(src, perm);
     // Output auto inizialitation if not yet initialized
-    auto_init_if_empty(*output->info(), input->info()->clone()->set_tensor_shape(output_shape));
+    auto_init_if_empty(*dst, src->clone()->set_tensor_shape(dst_shape));
 
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), output->info(), perm));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst, perm));
 
-    _input  = input;
-    _output = output;
-    _perm   = perm;
+    _perm = perm;
 
     // Create kernel
     CLBuildOptions build_opts;
-    build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(input->info()->data_type())));
-    build_opts.add_option("-DDEPTH_IN=" + support::cpp11::to_string(input->info()->dimension(2)));
+    build_opts.add_option("-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(data_size_from_type(src->data_type())));
+    build_opts.add_option("-DDEPTH_IN=" + support::cpp11::to_string(src->dimension(2)));
     // New positions of  width(W), height(H), channel(C) and batch(D) based on permutation vector
     build_opts.add_option("-DP1=" + support::cpp11::to_string((_perm.num_dimensions() >= 1) ? perm[0] : 0));
     build_opts.add_option("-DP2=" + support::cpp11::to_string((_perm.num_dimensions() >= 2) ? perm[1] : 1));
@@ -100,33 +107,36 @@ void CLPermuteKernel::configure(const CLCompileContext &compile_context, const I
     _kernel = create_kernel(compile_context, "permute", build_opts.options());
 
     // Configure  kernel window
-    Window win = calculate_max_window(*input->info(), Steps());
+    Window win = calculate_max_window(*src, Steps());
 
     // The CLPermute doesn't need padding so update_window_and_padding() can be skipped
     Coordinates coord;
-    coord.set_num_dimensions(output->info()->num_dimensions());
-    output->info()->set_valid_region(ValidRegion(coord, output->info()->tensor_shape()));
+    coord.set_num_dimensions(dst->num_dimensions());
+    dst->set_valid_region(ValidRegion(coord, dst->tensor_shape()));
 
     ICLKernel::configure_internal(win);
     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
 }
 
-Status CLPermuteKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
+Status ClPermuteKernel::validate(const ITensorInfo *src, const ITensorInfo *dst, const PermutationVector &perm)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output, perm));
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst, perm));
 
     return Status{};
 }
 
-void CLPermuteKernel::run(const Window &window, cl::CommandQueue &queue)
+void ClPermuteKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_MISMATCHING_WINDOWS(ICLKernel::window(), window);
 
+    const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
+    auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
+
     Window slice_in = window.first_slice_window_4D().collapse(ICLKernel::window(), 2, 4);
 
-    // Setup output slice
+    // Setup dst slice
     Window slice_out(slice_in);
     slice_out.set(Window::DimX, Window::Dimension(0, 0, 0));
     slice_out.set(Window::DimY, Window::Dimension(0, 0, 0));
@@ -136,10 +146,12 @@ void CLPermuteKernel::run(const Window &window, cl::CommandQueue &queue)
     do
     {
         unsigned int idx = 0;
-        add_4D_tensor_argument(idx, _input, slice_in);
-        add_4D_tensor_argument(idx, _output, slice_out);
+        add_4D_tensor_argument(idx, src, slice_in);
+        add_4D_tensor_argument(idx, dst, slice_out);
         enqueue(queue, *this, slice_in, lws_hint());
     }
     while(window.slide_window_slice_4D(slice_in) && window.slide_window_slice_4D(slice_out));
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/gpu/cl/kernels/ClPermuteKernel.h b/src/core/gpu/cl/kernels/ClPermuteKernel.h
new file mode 100644
index 0000000000..4cc72491bd
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClPermuteKernel.h
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2018-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_PERMUTE_KERNEL_H
+#define ARM_COMPUTE_CL_PERMUTE_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** OpenCL kernel to perform tensor permutation.
+ *
+ * Permutes given a permutation vector
+ */
+class ClPermuteKernel : public ICLKernel
+{
+public:
+    ClPermuteKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClPermuteKernel);
+    /** Set the src and dst of the kernel.
+     *
+     * @note Arbitrary permutation vectors are supported with rank not greater than 4
+     *
+     * @param[in] src  The src tensor info. Data types supported: All.
+     * @param[in] dst  The dst tensor info. Data types supported: Same as @p src
+     * @param[in] perm Permutation vector
+     */
+    void configure(const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm);
+    /** Set the src and dst of the kernel.
+     *
+     * @note Arbitrary permutation vectors are supported with rank not greater than 4
+     *
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] src             The src tensor info. Data types supported: All.
+     * @param[in] dst             The dst tensor info. Data types supported: Same as @p src
+     * @param[in] perm            Permutation vector
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm);
+    /** Static function to check if given info will lead to a valid configuration of @ref ClPermuteKernel
+     *
+     * @note Arbitrary permutation vectors are supported with rank not greater than 4
+     *
+     * @param[in] src  The src tensor info. Data types supported: All.
+     * @param[in] dst  The dst tensor info. Data types supported: same as @p src.
+     * @param[in] perm Permutation vector
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const PermutationVector &perm);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    PermutationVector _perm{};
+};
+} // namespace kernels
+} // namespace opencl
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CL_PERMUTE_KERNEL_H */
diff --git a/src/core/CL/kernels/CLReshapeLayerKernel.cpp b/src/core/gpu/cl/kernels/ClReshapeKernel.cpp
index 58d7843624..4da3fa0e03 100644
--- a/src/core/CL/kernels/CLReshapeLayerKernel.cpp
+++ b/src/core/gpu/cl/kernels/ClReshapeKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,7 +21,7 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include "src/core/CL/kernels/CLReshapeLayerKernel.h"
+#include "src/core/gpu/cl/kernels/ClReshapeKernel.h"
 
 #include "arm_compute/core/CL/CLHelpers.h"
 #include "arm_compute/core/CL/CLKernelLibrary.h"
@@ -38,73 +38,77 @@
 
 #include <string>
 
-/** [CLReshapeLayerKernel Kernel] **/
+/** [ClReshapeKernel Kernel] **/
 namespace arm_compute
 {
+namespace opencl
+{
+namespace kernels
+{
 namespace
 {
-Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output)
+Status validate_arguments(const ITensorInfo *src, const ITensorInfo *dst)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(input);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->data_type() == DataType::UNKNOWN);
+    ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_F16_UNSUPPORTED(src);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->data_type() == DataType::UNKNOWN);
 
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(input, output);
-    ARM_COMPUTE_RETURN_ERROR_ON(input->tensor_shape().total_size() != output->tensor_shape().total_size());
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_QUANTIZATION_INFO(src, dst);
+    ARM_COMPUTE_RETURN_ERROR_ON(src->tensor_shape().total_size() != dst->tensor_shape().total_size());
 
     return Status{};
 }
 } // namespace
 
-void CLReshapeLayerKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output)
+void ClReshapeKernel::configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
 {
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, output));
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(src, dst));
 
-    auto padding_info = get_padding_info({ input, output });
+    auto padding_info = get_padding_info({ src, dst });
 
     // Create kernel
-    std::set<std::string> build_opts = { "-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(input->element_size()) };
+    std::set<std::string> build_opts = { "-DDATA_TYPE=" + get_cl_unsigned_type_from_element_size(src->element_size()) };
     _kernel                          = create_kernel(compile_context, "reshape_layer", build_opts);
 
     // Add static arguments
-    const cl_int2 input_shape =
+    const cl_int2 src_shape =
     {
         {
-            static_cast<cl_int>(input->tensor_shape()[0]),
-            static_cast<cl_int>(input->tensor_shape()[1])
+            static_cast<cl_int>(src->tensor_shape()[0]),
+            static_cast<cl_int>(src->tensor_shape()[1])
         }
     };
-    const cl_int2 output_shape =
+    const cl_int2 dst_shape =
     {
         {
-            static_cast<cl_int>(output->tensor_shape()[0]),
-            static_cast<cl_int>(output->tensor_shape()[1])
+            static_cast<cl_int>(dst->tensor_shape()[0]),
+            static_cast<cl_int>(dst->tensor_shape()[1])
         }
     };
-    unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the input and output parameters
-    _kernel.setArg<cl_int2>(idx++, input_shape);
-    _kernel.setArg<cl_int2>(idx++, output_shape);
+    unsigned int idx = 2 * num_arguments_per_3D_tensor(); // Skip the src and dst parameters
+    _kernel.setArg<cl_int2>(idx++, src_shape);
+    _kernel.setArg<cl_int2>(idx++, dst_shape);
 
     // Configure kernel window
-    Window win = calculate_max_window(*input);
+    Window win = calculate_max_window(*src);
 
-    // Set the output valid region
-    output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape()));
+    // Set the dst valid region
+    dst->set_valid_region(ValidRegion(Coordinates(), dst->tensor_shape()));
     ICLKernel::configure_internal(win);
 
     ARM_COMPUTE_ERROR_ON(has_padding_changed(padding_info));
 }
 
-Status CLReshapeLayerKernel::validate(const ITensorInfo *input, const ITensorInfo *output)
+Status ClReshapeKernel::validate(const ITensorInfo *src, const ITensorInfo *dst)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(src, dst));
 
     return Status{};
 }
 
-void CLReshapeLayerKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
+void ClReshapeKernel::run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue)
 {
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
@@ -115,11 +119,13 @@ void CLReshapeLayerKernel::run_op(ITensorPack &tensors, const Window &window, cl
     const auto src = utils::cast::polymorphic_downcast<const ICLTensor *>(tensors.get_const_tensor(TensorType::ACL_SRC));
     auto       dst = utils::cast::polymorphic_downcast<ICLTensor *>(tensors.get_tensor(TensorType::ACL_DST));
 
-    // Set inputs
+    // Set srcs
     unsigned int idx = 0;
     add_3D_tensor_argument(idx, src, window_collapsed);
     add_3D_tensor_argument(idx, dst, window_collapsed);
     enqueue(queue, *this, slice, lws_hint());
 }
+} // namespace kernels
+} // namespace opencl
 } // namespace arm_compute
-/** [CLReshapeLayerKernel Kernel] **/
+/** [ClReshapeKernel Kernel] **/
diff --git a/src/core/gpu/cl/kernels/ClReshapeKernel.h b/src/core/gpu/cl/kernels/ClReshapeKernel.h
new file mode 100644
index 0000000000..ee835c0fd3
--- /dev/null
+++ b/src/core/gpu/cl/kernels/ClReshapeKernel.h
@@ -0,0 +1,66 @@
+/*
+ * Copyright (c) 2017-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_RESHAPE_KERNEL_H
+#define ARM_COMPUTE_CL_RESHAPE_KERNEL_H
+
+#include "src/core/common/Macros.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/IClKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+namespace kernels
+{
+/** Interface for the kernel to perform tensor reshaping */
+class ClReshapeKernel : public ICLKernel
+{
+public:
+    ClReshapeKernel() = default;
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(ClReshapeKernel);
+    /** Set the src and dst of the kernel
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info. Data type supported: All.
+     * @param[out] dst             Destination tensor info. Data type supported: Same as @p src
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref ClReshapeKernel
+     *
+     * @param[in] src Source tensor info. Data type supported: All
+     * @param[in] dst Destination tensor info. Data type supported: Same as @p src
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst);
+
+    // Inherited methods overridden:
+    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
+};
+} // namespace opencl
+} // namespace kernels
+} // namespace arm_compute
+#endif /*ARM_COMPUTE_CL_RESHAPE_KERNEL_H */
diff --git a/src/runtime/CL/functions/CLCopy.cpp b/src/runtime/CL/functions/CLCopy.cpp
index c3e30ada6e..98916bf38a 100644
--- a/src/runtime/CL/functions/CLCopy.cpp
+++ b/src/runtime/CL/functions/CLCopy.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,30 +23,58 @@
  */
 #include "arm_compute/runtime/CL/functions/CLCopy.h"
 
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/Error.h"
-#include "arm_compute/core/PixelValue.h"
-#include "arm_compute/core/TensorInfo.h"
+#include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClCopy.h"
 
 #include <utility>
 
-using namespace arm_compute;
+namespace arm_compute
+{
+struct CLCopy::Impl
+{
+    const ICLTensor                *src{ nullptr };
+    ICLTensor                      *dst{ nullptr };
+    std::unique_ptr<opencl::ClCopy> op{ nullptr };
+};
+
+CLCopy::CLCopy()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLCopy::CLCopy(CLCopy &&) = default;
+CLCopy &CLCopy::operator=(CLCopy &&) = default;
+CLCopy::~CLCopy()                    = default;
 
-void CLCopy::configure(ICLTensor *input, ICLTensor *output)
+void CLCopy::configure(ICLTensor *input, ICLTensor *output, Window *dst_window)
 {
-    configure(CLKernelLibrary::get().get_compile_context(), input, output);
+    configure(CLKernelLibrary::get().get_compile_context(), input, output, dst_window);
+}
+
+void CLCopy::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output, Window *dst_window)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input);
+
+    _impl->src = input;
+    _impl->dst = output;
+
+    _impl->op = std::make_unique<opencl::ClCopy>();
+    _impl->op->configure(compile_context, _impl->src->info(), _impl->dst->info(), dst_window);
 }
 
-void CLCopy::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *output)
+Status CLCopy::validate(const ITensorInfo *input, const ITensorInfo *output, Window *dst_window)
 {
-    auto k = std::make_unique<CLCopyKernel>();
-    k->configure(compile_context, input, output);
-    _kernel = std::move(k);
+    return opencl::ClCopy::validate(input, output, dst_window);
 }
 
-Status CLCopy::validate(const arm_compute::ITensorInfo *input, const arm_compute::ITensorInfo *output)
+void CLCopy::run()
 {
-    return CLCopyKernel::validate(input, output);
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+    _impl->op->run(pack);
 }
+} // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLCrop.cpp b/src/runtime/CL/functions/CLCrop.cpp
new file mode 100644
index 0000000000..20cab4df5f
--- /dev/null
+++ b/src/runtime/CL/functions/CLCrop.cpp
@@ -0,0 +1,82 @@
+/*
+ * Copyright (c) 2018-2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/runtime/CL/functions/CLCrop.h"
+
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClCrop.h"
+
+#include <utility>
+
+namespace arm_compute
+{
+struct CLCrop::Impl
+{
+    const ICLTensor                *src{ nullptr };
+    ICLTensor                      *dst{ nullptr };
+    std::unique_ptr<opencl::ClCrop> op{ nullptr };
+};
+
+CLCrop::CLCrop()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLCrop::CLCrop(CLCrop &&) = default;
+CLCrop &CLCrop::operator=(CLCrop &&) = default;
+CLCrop::~CLCrop()                    = default;
+
+void CLCrop::configure(const ICLTensor *src, ICLTensor *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value,
+                       Window *dst_window)
+{
+    configure(CLKernelLibrary::get().get_compile_context(), src, dst, start, end, batch_index, extrapolation_value, dst_window);
+}
+
+void CLCrop::configure(const CLCompileContext &compile_context, const ICLTensor *src, ICLTensor *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value,
+                       Window *dst_window)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(src, dst);
+
+    _impl->src = src;
+    _impl->dst = dst;
+
+    _impl->op = std::make_unique<opencl::ClCrop>();
+    _impl->op->configure(compile_context, _impl->src->info(), _impl->dst->info(), start, end, batch_index, extrapolation_value, dst_window);
+}
+
+Status CLCrop::validate(const ITensorInfo *input, const ITensorInfo *output, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *dst_window)
+{
+    return opencl::ClCrop::validate(input, output, start, end, batch_index, extrapolation_value, dst_window);
+}
+
+void CLCrop::run()
+{
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+    _impl->op->run(pack);
+}
+} // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLCropResize.cpp b/src/runtime/CL/functions/CLCropResize.cpp
index ed31446cf9..77c44d539b 100644
--- a/src/runtime/CL/functions/CLCropResize.cpp
+++ b/src/runtime/CL/functions/CLCropResize.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -25,10 +25,7 @@
 
 #include "arm_compute/core/CL/CLHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
-#include "src/core/CL/kernels/CLCropKernel.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
 #include "src/core/helpers/WindowHelpers.h"
 
@@ -59,7 +56,7 @@ inline void configure_crop(const ICLTensor *input, ICLTensor *crop_boxes, ICLTen
 } // namespace
 
 CLCropResize::CLCropResize()
-    : _input(nullptr), _boxes(nullptr), _box_ind(nullptr), _output(nullptr), _num_boxes(0), _method(), _extrapolation_value(0), _scale(), _copy(), _crop_results(), _scaled_results(), _internal_kernels()
+    : _input(nullptr), _boxes(nullptr), _box_ind(nullptr), _output(nullptr), _num_boxes(0), _method(), _extrapolation_value(0), _scale(), _copy(), _crop_results(), _scaled_results(), _internal_functions()
 {
 }
 
@@ -73,7 +70,7 @@ Status CLCropResize::validate(const ITensorInfo *input, ITensorInfo *boxes, ITen
     ARM_COMPUTE_RETURN_ERROR_ON(boxes->tensor_shape()[0] != 4);
     ARM_COMPUTE_RETURN_ERROR_ON(boxes->tensor_shape()[1] != box_ind->tensor_shape()[0]);
     TensorInfo temp_info;
-    ARM_COMPUTE_RETURN_ON_ERROR(CLCropKernel::validate(input->clone().get(), &temp_info, { 0, 0 }, { 1, 1 }, input->dimension(3) - 1, extrapolation_value));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLCrop::validate(input->clone().get(), &temp_info, { 0, 0 }, { 1, 1 }, input->dimension(3) - 1, extrapolation_value));
     if(output->total_size() > 0)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_NOT_IN(output, DataType::F32);
@@ -111,11 +108,11 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
 
     // For each crop box:
     // - The initial cropped image is produced as specified by boxes[i] from the 3D image input[box_ind[i]].
-    //   Possibly using a CLCropKernel and up to four CLMemsetKernels.
+    //   Possibly using a CLCrop and up to four CLFills.
     // - A tensor is required to hold this initial cropped image.
     // - A scale function is used to resize the cropped image to the size specified by crop_size.
     // - A tensor is required to hold the final scaled image before it is copied into the 4D output
-    //   that will hold all final cropped and scaled 3D images using CLCopyKernel.
+    //   that will hold all final cropped and scaled 3D images using CLCopy.
 
     // The contents of _boxes and _box_ind are required to calculate the shape
     // of the initial cropped image and thus are required to configure the
@@ -149,7 +146,7 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
         Window win = calculate_max_window(*_output->info());
         win.set(3, Window::Dimension(num_box, num_box + 1, 1));
 
-        auto copy_kernel = std::make_unique<CLCopyKernel>();
+        auto copy_kernel = std::make_unique<CLCopy>();
         copy_kernel->configure(compile_context, _scaled_results[num_box].get(), _output, &win);
         _copy.emplace_back(std::move(copy_kernel));
 
@@ -207,9 +204,10 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
         {
             Window slice_fill_rows_before(full_window);
             slice_fill_rows_before.set(2, Window::Dimension(0, rows_out_of_bounds[0], 1));
-            auto kernel = std::make_unique<CLMemsetKernel>();
+            auto kernel = std::make_unique<CLFill>();
             kernel->configure(compile_context, _crop_results[num_box].get(), extrapolation_value, &slice_fill_rows_before);
-            _internal_kernels.push_back(std::move(kernel));
+            //_internal_functions.emplace_back(std::move(kernel));
+            _internal_functions.push_back(std::move(kernel));
         }
 
         Window slice_in(full_window);
@@ -224,18 +222,20 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
             {
                 Window slice_fill_cols_before(slice_in);
                 slice_fill_cols_before.set(1, Window::Dimension(0, cols_out_of_bounds[0], 1));
-                auto kernel = std::make_unique<CLMemsetKernel>();
+                auto kernel = std::make_unique<CLFill>();
                 kernel->configure(compile_context, _crop_results[num_box].get(), extrapolation_value, &slice_fill_cols_before);
-                _internal_kernels.push_back(std::move(kernel));
+                //_internal_functions.emplace_back(std::move(kernel));
+                _internal_functions.push_back(std::move(kernel));
             }
 
             if(cols_out_of_bounds[1] > 0)
             {
                 Window slice_fill_cols_after(slice_in);
                 slice_fill_cols_after.set(1, Window::Dimension(_crop_results[num_box].get()->info()->dimension(1) - cols_out_of_bounds[1], _crop_results[num_box].get()->info()->dimension(1), 1));
-                auto kernel = std::make_unique<CLMemsetKernel>();
+                auto kernel = std::make_unique<CLFill>();
                 kernel->configure(compile_context, _crop_results[num_box].get(), extrapolation_value, &slice_fill_cols_after);
-                _internal_kernels.push_back(std::move(kernel));
+                //_internal_functions.emplace_back(std::move(kernel));
+                _internal_functions.push_back(std::move(kernel));
             }
 
             // Copy all elements within the input bounds from the input tensor.
@@ -246,10 +246,11 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
                                         is_height_flipped ? start[1] - rows_out_of_bounds[0] : start[1] + rows_out_of_bounds[0] };
                 Coordinates2D end_in{ is_width_flipped ? start_in.x - cols_in_bounds + 1 : start_in.x + cols_in_bounds - 1,
                                       is_height_flipped ? start_in.y - rows_in_bounds + 1 : start_in.y + rows_in_bounds - 1 };
-                auto kernel = std::make_unique<CLCropKernel>();
+                auto kernel = std::make_unique<CLCrop>();
 
                 kernel->configure(compile_context, _input, _crop_results[num_box].get(), start_in, end_in, batch_index, extrapolation_value, &slice_in);
-                _internal_kernels.push_back(std::move(kernel));
+                //_internal_functions.emplace_back(std::move(kernel));
+                _internal_functions.push_back(std::move(kernel));
             }
         }
 
@@ -258,9 +259,10 @@ void CLCropResize::configure(const CLCompileContext &compile_context, const ICLT
         {
             Window slice_fill_rows_after(full_window);
             slice_fill_rows_after.set(2, Window::Dimension(_crop_results[num_box].get()->info()->dimension(2) - rows_out_of_bounds[1], _crop_results[num_box].get()->info()->dimension(2), 1));
-            auto kernel = std::make_unique<CLMemsetKernel>();
+            auto kernel = std::make_unique<CLFill>();
             kernel->configure(compile_context, _crop_results[num_box].get(), extrapolation_value, &slice_fill_rows_after);
-            _internal_kernels.push_back(std::move(kernel));
+            //_internal_functions.emplace_back(std::move(kernel));
+            _internal_functions.push_back(std::move(kernel));
         }
     }
     _boxes->unmap(CLScheduler::get().queue());
@@ -272,9 +274,9 @@ void CLCropResize::run()
 {
     ARM_COMPUTE_ERROR_ON_MSG(_output == nullptr, "Unconfigured function");
 
-    for(unsigned int i = 0; i < _internal_kernels.size(); ++i)
+    for(unsigned int i = 0; i < _internal_functions.size(); ++i)
     {
-        CLScheduler::get().enqueue(*(_internal_kernels[i]));
+        _internal_functions[i]->run();
     }
 
     CLScheduler::get().sync();
@@ -285,7 +287,7 @@ void CLCropResize::run()
     CLScheduler::get().sync();
     for(auto &kernel : _copy)
     {
-        CLScheduler::get().enqueue(*kernel, true);
+        kernel->run();
     }
     CLScheduler::get().sync();
 }
diff --git a/src/runtime/CL/functions/CLDeconvolutionLayerUpsample.cpp b/src/runtime/CL/functions/CLDeconvolutionLayerUpsample.cpp
index 4989f6460d..c371558f30 100644
--- a/src/runtime/CL/functions/CLDeconvolutionLayerUpsample.cpp
+++ b/src/runtime/CL/functions/CLDeconvolutionLayerUpsample.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,13 +28,12 @@
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
 #include "src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 
 namespace arm_compute
 {
 CLDeconvolutionLayerUpsample::CLDeconvolutionLayerUpsample() // NOLINT
     : _upsample(std::make_unique<CLDeconvolutionLayerUpsampleKernel>()),
-      _memset(std::make_unique<CLMemsetKernel>()),
+      _fill(),
       _output(nullptr)
 {
 }
@@ -56,13 +55,13 @@ void CLDeconvolutionLayerUpsample::configure(const CLCompileContext &compile_con
     ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
     _output = output;
-    _memset->configure(compile_context, _output, PixelValue(0, _output->info()->data_type(), _output->info()->quantization_info()));
+    _fill.configure(compile_context, _output, PixelValue(0, _output->info()->data_type(), _output->info()->quantization_info()));
     _upsample->configure(compile_context, input, _output, info);
 }
 
 void CLDeconvolutionLayerUpsample::run()
 {
-    CLScheduler::get().enqueue(*_memset, false);
+    _fill.run();
     CLScheduler::get().enqueue(*_upsample, true);
 }
 } // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLDirectDeconvolutionLayer.cpp b/src/runtime/CL/functions/CLDirectDeconvolutionLayer.cpp
index 0e3109439e..d802ef2dc9 100644
--- a/src/runtime/CL/functions/CLDirectDeconvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLDirectDeconvolutionLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -32,7 +32,6 @@
 #include "src/core/CL/kernels/CLDeconvolutionLayerUpsampleKernel.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
 #include "src/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 #include "src/core/CL/kernels/CLWeightsReshapeKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
 
diff --git a/src/runtime/CL/functions/CLFFTConvolutionLayer.cpp b/src/runtime/CL/functions/CLFFTConvolutionLayer.cpp
index 45e74df703..41b02d03f2 100644
--- a/src/runtime/CL/functions/CLFFTConvolutionLayer.cpp
+++ b/src/runtime/CL/functions/CLFFTConvolutionLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,7 +29,6 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CPP/CPPScheduler.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
 #include "src/core/CL/kernels/CLFFTDigitReverseKernel.h"
 #include "src/core/CL/kernels/CLFFTRadixStageKernel.h"
 #include "src/core/CL/kernels/CLFFTScaleKernel.h"
diff --git a/src/runtime/CL/functions/CLFill.cpp b/src/runtime/CL/functions/CLFill.cpp
index 30843d8cc0..b22d79fea4 100644
--- a/src/runtime/CL/functions/CLFill.cpp
+++ b/src/runtime/CL/functions/CLFill.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,22 +23,56 @@
  */
 #include "arm_compute/runtime/CL/functions/CLFill.h"
 
+#include "arm_compute/core/CL/CLKernelLibrary.h"
+#include "arm_compute/core/CL/ICLTensor.h"
 #include "arm_compute/core/Types.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClFill.h"
 
 #include <utility>
 
 namespace arm_compute
 {
-void CLFill::configure(ICLTensor *tensor, PixelValue constant_value)
+struct CLFill::Impl
 {
-    configure(CLKernelLibrary::get().get_compile_context(), tensor, constant_value);
+    const ICLTensor                *src{ nullptr };
+    ICLTensor                      *dst{ nullptr };
+    std::unique_ptr<opencl::ClFill> op{ nullptr };
+};
+
+CLFill::CLFill()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLFill::CLFill(CLFill &&) = default;
+CLFill &CLFill::operator=(CLFill &&) = default;
+CLFill::~CLFill()                    = default;
+
+void CLFill::configure(ICLTensor *tensor, const PixelValue &constant_value, Window *dst_window)
+{
+    configure(CLKernelLibrary::get().get_compile_context(), tensor, constant_value, dst_window);
+}
+
+void CLFill::configure(const CLCompileContext &compile_context, ICLTensor *tensor, const PixelValue &constant_value, Window *dst_window)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(tensor);
+
+    _impl->src = tensor;
+
+    _impl->op = std::make_unique<opencl::ClFill>();
+    _impl->op->configure(compile_context, _impl->src->info(), constant_value, dst_window);
+}
+
+Status CLFill::validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *dst_window)
+{
+    return opencl::ClFill::validate(tensor, constant_value, dst_window);
 }
 
-void CLFill::configure(const CLCompileContext &compile_context, ICLTensor *tensor, PixelValue constant_value)
+void CLFill::run()
 {
-    auto k = std::make_unique<CLMemsetKernel>();
-    k->configure(compile_context, tensor, constant_value);
-    _kernel = std::move(k);
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    _impl->op->run(pack);
 }
 } // namespace arm_compute
diff --git a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
index e536816f97..365f95243f 100644
--- a/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
+++ b/src/runtime/CL/functions/CLGenerateProposalsLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,7 +29,6 @@
 #include "src/core/CL/kernels/CLDequantizationLayerKernel.h"
 #include "src/core/CL/kernels/CLGenerateProposalsLayerKernel.h"
 #include "src/core/CL/kernels/CLPadLayerKernel.h"
-#include "src/core/CL/kernels/CLPermuteKernel.h"
 #include "src/core/CL/kernels/CLQuantizationLayerKernel.h"
 #include "src/core/helpers/AutoConfiguration.h"
 
@@ -37,9 +36,9 @@ namespace arm_compute
 {
 CLGenerateProposalsLayer::CLGenerateProposalsLayer(std::shared_ptr<IMemoryManager> memory_manager)
     : _memory_group(memory_manager),
-      _permute_deltas_kernel(std::make_unique<CLPermuteKernel>()),
+      _permute_deltas(),
       _flatten_deltas(),
-      _permute_scores_kernel(std::make_unique<CLPermuteKernel>()),
+      _permute_scores(),
       _flatten_scores(),
       _compute_anchors_kernel(std::make_unique<CLComputeAllAnchorsKernel>()),
       _bounding_box_kernel(std::make_unique<CLBoundingBoxTransformKernel>()),
@@ -110,7 +109,7 @@ void CLGenerateProposalsLayer::configure(const CLCompileContext &compile_context
     if(!_is_nhwc)
     {
         _memory_group.manage(&_deltas_permuted);
-        _permute_deltas_kernel->configure(compile_context, deltas, &_deltas_permuted, PermutationVector{ 2, 0, 1 });
+        _permute_deltas.configure(compile_context, deltas, &_deltas_permuted, PermutationVector{ 2, 0, 1 });
         _flatten_deltas.configure(compile_context, &_deltas_permuted, &_deltas_flattened);
         _deltas_permuted.allocator()->allocate();
     }
@@ -127,7 +126,7 @@ void CLGenerateProposalsLayer::configure(const CLCompileContext &compile_context
     if(!_is_nhwc)
     {
         _memory_group.manage(&_scores_permuted);
-        _permute_scores_kernel->configure(compile_context, scores, &_scores_permuted, PermutationVector{ 2, 0, 1 });
+        _permute_scores.configure(compile_context, scores, &_scores_permuted, PermutationVector{ 2, 0, 1 });
         _flatten_scores.configure(compile_context, &_scores_permuted, &_scores_flattened);
         _scores_permuted.allocator()->allocate();
     }
@@ -244,8 +243,8 @@ Status CLGenerateProposalsLayer::validate(const ITensorInfo *scores, const ITens
     }
     else
     {
-        ARM_COMPUTE_RETURN_ON_ERROR(CLPermuteKernel::validate(deltas, &deltas_permuted_info, PermutationVector{ 2, 0, 1 }));
-        ARM_COMPUTE_RETURN_ON_ERROR(CLPermuteKernel::validate(scores, &scores_permuted_info, PermutationVector{ 2, 0, 1 }));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLPermute::validate(deltas, &deltas_permuted_info, PermutationVector{ 2, 0, 1 }));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLPermute::validate(scores, &scores_permuted_info, PermutationVector{ 2, 0, 1 }));
     }
 
     TensorInfo deltas_flattened_info(deltas->clone()->set_tensor_shape(TensorShape(values_per_roi, total_num_anchors)).set_is_resizable(true));
@@ -356,8 +355,8 @@ void CLGenerateProposalsLayer::run()
     // Transpose and reshape the inputs
     if(!_is_nhwc)
     {
-        CLScheduler::get().enqueue(*_permute_deltas_kernel, false);
-        CLScheduler::get().enqueue(*_permute_scores_kernel, false);
+        _permute_deltas.run();
+        _permute_scores.run();
     }
     _flatten_deltas.run();
     _flatten_scores.run();
diff --git a/src/runtime/CL/functions/CLLSTMLayer.cpp b/src/runtime/CL/functions/CLLSTMLayer.cpp
index 77df917119..5036126aea 100644
--- a/src/runtime/CL/functions/CLLSTMLayer.cpp
+++ b/src/runtime/CL/functions/CLLSTMLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,7 +29,6 @@
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/core/utils/quantization/AsymmHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
 #include "src/core/CL/kernels/CLDepthConvertLayerKernel.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
 #include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h"
@@ -42,7 +41,6 @@
 #include "src/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h"
 #include "src/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
 #include "src/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 #include "src/core/CL/kernels/CLTransposeKernel.h"
 
 namespace arm_compute
@@ -55,15 +53,14 @@ CLLSTMLayer::CLLSTMLayer(std::shared_ptr<IMemoryManager> memory_manager)
       _fully_connected_forget_gate(), _accum_forget_gate1(), _pixelwise_mul_forget_gate(), _activation_forget_gate(), _fully_connected_cell_state(), _gemm_cell_state1(),
       _transpose_cell_state(std::make_unique<CLTransposeKernel>()), _accum_cell_state1(), _accum_cell_state2(), _pixelwise_mul_cell_state1(), _activation_cell_state(), _cell_clip(),
       _pixelwise_mul_cell_state2(), _fully_connected_output(), _pixelwise_mul_output_state1(), _accum_output1(), _activation_output(), _activation_output_state(), _pixelwise_mul_output_state2(),
-      _fully_connected_output_state(), _projection_clip(), _copy_cell_state(std::make_unique<CLCopyKernel>()), _copy_output(std::make_unique<CLCopyKernel>()), _concat_scratch_buffer(),
-      _concat_inputs_forget_gate(), _concat_weights_forget_gate(), _concat_weights_input_gate(), _concat_weights_output(), _ones_memset_kernel(std::make_unique<CLMemsetKernel>()),
-      _mean_std_norm_input_gate(), _pixelwise_mul_input_gate_coeff(), _accum_input_gate_bias(), _mean_std_norm_forget_gate(), _pixelwise_mul_forget_gate_coeff(), _accum_forget_gate_bias(),
-      _mean_std_norm_cell_gate(), _pixelwise_mul_cell_gate_coeff(), _accum_cell_gate_bias(), _mean_std_norm_output_gate(), _pixelwise_mul_output_gate_coeff(), _accum_output_gate_bias(), _input_gate_out1(),
-      _input_gate_out2(), _input_gate_out3(), _input_gate_out4(), _forget_gate_out1(), _forget_gate_out2(), _forget_gate_out3(), _forget_gate_out4(), _forget_gate_out5(), _forget_gate_out6(),
-      _cell_state_out1(), _cell_state_out2(), _cell_state_out3(), _cell_state_out4(), _cell_state_out5(), _output1(), _output2(), _output3(), _output4(), _cell_state_activation(), _output_state1(), _ones(),
-      _input_layer_norm_out1(), _input_layer_norm_out2(), _forget_layer_norm_out1(), _forget_layer_norm_out2(), _cell_layer_norm_out1(), _cell_layer_norm_out2(), _output_layer_norm_out1(),
-      _output_layer_norm_out2(), _run_peephole_opt(false), _run_cifg_opt(false), _perform_cell_clipping(false), _has_projection_weights(false), _perform_projection_clipping(false), _is_prepared(false),
-      _is_layer_norm_lstm(false)
+      _fully_connected_output_state(), _projection_clip(), _copy_cell_state(), _copy_output(), _concat_scratch_buffer(), _concat_inputs_forget_gate(), _concat_weights_forget_gate(),
+      _concat_weights_input_gate(), _concat_weights_output(), _ones_fill(), _mean_std_norm_input_gate(), _pixelwise_mul_input_gate_coeff(), _accum_input_gate_bias(), _mean_std_norm_forget_gate(),
+      _pixelwise_mul_forget_gate_coeff(), _accum_forget_gate_bias(), _mean_std_norm_cell_gate(), _pixelwise_mul_cell_gate_coeff(), _accum_cell_gate_bias(), _mean_std_norm_output_gate(),
+      _pixelwise_mul_output_gate_coeff(), _accum_output_gate_bias(), _input_gate_out1(), _input_gate_out2(), _input_gate_out3(), _input_gate_out4(), _forget_gate_out1(), _forget_gate_out2(),
+      _forget_gate_out3(), _forget_gate_out4(), _forget_gate_out5(), _forget_gate_out6(), _cell_state_out1(), _cell_state_out2(), _cell_state_out3(), _cell_state_out4(), _cell_state_out5(), _output1(),
+      _output2(), _output3(), _output4(), _cell_state_activation(), _output_state1(), _ones(), _input_layer_norm_out1(), _input_layer_norm_out2(), _forget_layer_norm_out1(), _forget_layer_norm_out2(),
+      _cell_layer_norm_out1(), _cell_layer_norm_out2(), _output_layer_norm_out1(), _output_layer_norm_out2(), _run_peephole_opt(false), _run_cifg_opt(false), _perform_cell_clipping(false),
+      _has_projection_weights(false), _perform_projection_clipping(false), _is_prepared(false), _is_layer_norm_lstm(false)
 {
 }
 
@@ -190,7 +187,7 @@ void CLLSTMLayer::configure(const CLCompileContext &compile_context, const ICLTe
     {
         _memory_group.manage(&_input_gate_out1);
         _ones.allocator()->init(TensorInfo(cell_state_shape, 1, input->info()->data_type()));
-        _ones_memset_kernel->configure(compile_context, &_ones, PixelValue(1, _ones.info()->data_type()));
+        _ones_fill.configure(compile_context, &_ones, PixelValue(1, _ones.info()->data_type()));
         _subtract_input_gate.configure(compile_context, &_ones, forget_gate_out, &_input_gate_out1, ConvertPolicy::SATURATE);
         _ones.allocator()->allocate();
         _run_cifg_opt = true;
@@ -385,8 +382,8 @@ void CLLSTMLayer::configure(const CLCompileContext &compile_context, const ICLTe
     }
 
     // Copy cell state and output
-    _copy_cell_state->configure(compile_context, &_cell_state_out1, cell_state_out);
-    _copy_output->configure(compile_context, output_state_out, output);
+    _copy_cell_state.configure(compile_context, &_cell_state_out1, cell_state_out);
+    _copy_output.configure(compile_context, output_state_out, output);
 
     // Vector for holding the tensors to store in scratch buffer
     std::vector<const ICLTensor *> scratch_inputs;
@@ -618,8 +615,8 @@ Status CLLSTMLayer::validate(const ITensorInfo *input,
     }
 
     // Validate copy kernel
-    ARM_COMPUTE_RETURN_ON_ERROR(CLCopyKernel::validate(&cell_state_tmp, cell_state_out));
-    ARM_COMPUTE_RETURN_ON_ERROR(CLCopyKernel::validate(output_state_out, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLCopy::validate(&cell_state_tmp, cell_state_out));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLCopy::validate(output_state_out, output));
 
     // Validate scratch concatenation
     std::vector<const ITensorInfo *> inputs_vector_info_raw;
@@ -660,7 +657,7 @@ void CLLSTMLayer::run()
 
     if(_run_cifg_opt)
     {
-        CLScheduler::get().enqueue(*_ones_memset_kernel);
+        _ones_fill.run();
         _subtract_input_gate.run();
     }
     else
@@ -729,8 +726,8 @@ void CLLSTMLayer::run()
         }
     }
 
-    CLScheduler::get().enqueue(*_copy_cell_state);
-    CLScheduler::get().enqueue(*_copy_output);
+    _copy_cell_state.run();
+    _copy_output.run();
 
     _concat_scratch_buffer.run();
 }
diff --git a/src/runtime/CL/functions/CLMaxUnpoolingLayer.cpp b/src/runtime/CL/functions/CLMaxUnpoolingLayer.cpp
index c9deb301ef..52151cdfe1 100644
--- a/src/runtime/CL/functions/CLMaxUnpoolingLayer.cpp
+++ b/src/runtime/CL/functions/CLMaxUnpoolingLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,12 +28,11 @@
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "src/core/CL/kernels/CLMaxUnpoolingLayerKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 
 namespace arm_compute
 {
 CLMaxUnpoolingLayer::CLMaxUnpoolingLayer()
-    : _memset_kernel(std::make_unique<CLMemsetKernel>()),
+    : _fill(),
       _unpooling_layer_kernel(std::make_unique<CLMaxUnpoolingLayerKernel>())
 {
 }
@@ -48,7 +47,7 @@ void CLMaxUnpoolingLayer::configure(ICLTensor *input, ICLTensor *indices, ICLTen
 void CLMaxUnpoolingLayer::configure(const CLCompileContext &compile_context, ICLTensor *input, ICLTensor *indices, ICLTensor *output, const PoolingLayerInfo &pool_info)
 {
     const PixelValue zero_value(0.f);
-    _memset_kernel->configure(output, zero_value);
+    _fill.configure(output, zero_value);
 
     _unpooling_layer_kernel->configure(compile_context, input, indices, output, pool_info);
 }
@@ -60,8 +59,8 @@ Status CLMaxUnpoolingLayer::validate(const ITensorInfo *input, const ITensorInfo
 
 void CLMaxUnpoolingLayer::run()
 {
-    // Run memset
-    CLScheduler::get().enqueue(*_memset_kernel, false);
+    // Run fill
+    _fill.run();
 
     // Run max unpooling layer
     CLScheduler::get().enqueue(*_unpooling_layer_kernel);
diff --git a/src/runtime/CL/functions/CLPadLayer.cpp b/src/runtime/CL/functions/CLPadLayer.cpp
index 8c5d529117..d105c0597c 100644
--- a/src/runtime/CL/functions/CLPadLayer.cpp
+++ b/src/runtime/CL/functions/CLPadLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -22,14 +22,13 @@
  * SOFTWARE.
  */
 #include "arm_compute/runtime/CL/functions/CLPadLayer.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
 #include "src/core/CL/kernels/CLPadLayerKernel.h"
 
 namespace arm_compute
 {
 CLPadLayer::CLPadLayer()
     : _pad_kernel(std::make_unique<CLPadLayerKernel>()),
-      _copy_kernel(std::make_unique<CLCopyKernel>()),
+      _copy(),
       _perform_pad(false)
 {
 }
@@ -57,7 +56,7 @@ void CLPadLayer::configure(const CLCompileContext &compile_context, ICLTensor *i
     else
     {
         // Copy the input to the whole output if no padding is applied
-        _copy_kernel->configure(compile_context, input, output);
+        _copy.configure(compile_context, input, output);
     }
 }
 Status CLPadLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const PaddingList &padding, PixelValue constant_value, PaddingMode mode)
@@ -73,7 +72,7 @@ Status CLPadLayer::validate(const ITensorInfo *input, const ITensorInfo *output,
     }
     else
     {
-        ARM_COMPUTE_RETURN_ON_ERROR(CLCopyKernel::validate(input, output));
+        ARM_COMPUTE_RETURN_ON_ERROR(CLCopy::validate(input, output));
     }
     return Status{};
 }
@@ -85,7 +84,7 @@ void CLPadLayer::run()
     }
     else
     {
-        CLScheduler::get().enqueue(*_copy_kernel);
+        _copy.run();
     }
 }
 } // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/CL/functions/CLPermute.cpp b/src/runtime/CL/functions/CLPermute.cpp
index 31b152c553..eeb0169241 100644
--- a/src/runtime/CL/functions/CLPermute.cpp
+++ b/src/runtime/CL/functions/CLPermute.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,12 +23,30 @@
  */
 #include "arm_compute/runtime/CL/functions/CLPermute.h"
 
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "arm_compute/core/Error.h"
-#include "src/core/CL/kernels/CLPermuteKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClPermute.h"
 
 namespace arm_compute
 {
+struct CLPermute::Impl
+{
+    const ICLTensor                   *src{ nullptr };
+    ICLTensor                         *dst{ nullptr };
+    std::unique_ptr<opencl::ClPermute> op{ nullptr };
+};
+
+CLPermute::CLPermute()
+    : _impl(std::make_unique<Impl>())
+{
+}
+CLPermute::CLPermute(CLPermute &&) = default;
+CLPermute &CLPermute::operator=(CLPermute &&) = default;
+CLPermute::~CLPermute()                       = default;
+
 void CLPermute::configure(const ICLTensor *input, ICLTensor *output, const PermutationVector &perm)
 {
     configure(CLKernelLibrary::get().get_compile_context(), input, output, perm);
@@ -36,14 +54,25 @@ void CLPermute::configure(const ICLTensor *input, ICLTensor *output, const Permu
 
 void CLPermute::configure(const CLCompileContext &compile_context, const ICLTensor *input, ICLTensor *output, const PermutationVector &perm)
 {
-    auto k = std::make_unique<CLPermuteKernel>();
-    k->configure(compile_context, input, output, perm);
-    _kernel = std::move(k);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
+
+    _impl->src = input;
+    _impl->dst = output;
+
+    _impl->op = std::make_unique<opencl::ClPermute>();
+    _impl->op->configure(compile_context, _impl->src->info(), _impl->dst->info(), perm);
 }
 
 Status CLPermute::validate(const ITensorInfo *input, const ITensorInfo *output, const PermutationVector &perm)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(CLPermuteKernel::validate(input, output, perm));
-    return Status{};
+    return opencl::ClPermute::validate(input, output, perm);
+}
+
+void CLPermute::run()
+{
+    ITensorPack pack;
+    pack.add_tensor(TensorType::ACL_SRC, _impl->src);
+    pack.add_tensor(TensorType::ACL_DST, _impl->dst);
+    _impl->op->run(pack);
 }
-} // namespace arm_compute
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/CL/functions/CLQLSTMLayer.cpp b/src/runtime/CL/functions/CLQLSTMLayer.cpp
index 4395a39060..e7a0e5765e 100644
--- a/src/runtime/CL/functions/CLQLSTMLayer.cpp
+++ b/src/runtime/CL/functions/CLQLSTMLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -30,7 +30,6 @@
 #include "arm_compute/core/utils/misc/InfoHelpers.h"
 #include "arm_compute/core/utils/quantization/AsymmHelpers.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
 #include "src/core/CL/kernels/CLDepthConvertLayerKernel.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
 #include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h"
@@ -106,7 +105,7 @@ CLQLSTMLayer::CLQLSTMLayer(std::shared_ptr<IMemoryManager> memory_manager)
       _recurrent_to_output_reduction(std::make_unique<CLGEMMLowpMatrixAReductionKernel>()),
       _projection_reduction(std::make_unique<CLGEMMLowpMatrixAReductionKernel>()),
       _layer_norms(),
-      _copy_output(std::make_unique<CLCopyKernel>())
+      _copy_output()
 {
     for(auto &norm : _layer_norms)
     {
@@ -593,7 +592,7 @@ void CLQLSTMLayer::configure(const CLCompileContext &compile_context, const ICLT
     }
 
     // Copy output_state_out to output
-    _copy_output->configure(compile_context, output_state_out, output);
+    _copy_output.configure(compile_context, output_state_out, output);
 }
 
 Status CLQLSTMLayer::validate(const ITensorInfo *input,
@@ -957,7 +956,7 @@ Status CLQLSTMLayer::validate(const ITensorInfo *input,
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(output_state_in, output_state_out);
     }
 
-    ARM_COMPUTE_RETURN_ON_ERROR(CLCopyKernel::validate(output_state_out, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLCopy::validate(output_state_out, output));
     return Status{};
 }
 
@@ -1099,7 +1098,7 @@ void CLQLSTMLayer::run()
     }
 
     // Copy output_state_out to output
-    CLScheduler::get().enqueue(*_copy_output);
+    _copy_output.run();
 }
 
 void CLQLSTMLayer::prepare()
diff --git a/src/runtime/CL/functions/CLRNNLayer.cpp b/src/runtime/CL/functions/CLRNNLayer.cpp
index 2a99ece388..967f4aa41b 100644
--- a/src/runtime/CL/functions/CLRNNLayer.cpp
+++ b/src/runtime/CL/functions/CLRNNLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -28,7 +28,6 @@
 #include "arm_compute/core/Utils.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLCopyKernel.h"
 #include "src/core/CL/kernels/CLDepthConvertLayerKernel.h"
 #include "src/core/CL/kernels/CLFillBorderKernel.h"
 #include "src/core/CL/kernels/CLGEMMLowpMatrixMultiplyNativeKernel.h"
@@ -47,8 +46,8 @@ namespace arm_compute
 using namespace arm_compute::misc::shape_calculator;
 
 CLRNNLayer::CLRNNLayer(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_group(std::move(memory_manager)), _gemm_state_f(), _add_kernel(), _activation(), _fully_connected_kernel(), _copy_kernel(std::make_unique<CLCopyKernel>()), _fully_connected_out(),
-      _gemm_output(), _add_output(), _is_prepared(false)
+    : _memory_group(std::move(memory_manager)), _gemm_state_f(), _add_kernel(), _activation(), _fully_connected_kernel(), _copy(), _fully_connected_out(), _gemm_output(), _add_output(),
+      _is_prepared(false)
 {
 }
 
@@ -122,7 +121,7 @@ void CLRNNLayer::configure(const CLCompileContext &compile_context, const ICLTen
     _activation.configure(compile_context, &_add_output, hidden_state, info);
     _add_output.allocator()->allocate();
 
-    _copy_kernel->configure(compile_context, hidden_state, output);
+    _copy.configure(compile_context, hidden_state, output);
 }
 
 void CLRNNLayer::run()
@@ -137,7 +136,7 @@ void CLRNNLayer::run()
     _activation.run();
 
     // copy hidden out to output
-    CLScheduler::get().enqueue(*_copy_kernel);
+    _copy.run();
 }
 
 void CLRNNLayer::prepare()
diff --git a/src/runtime/CL/functions/CLReshapeLayer.cpp b/src/runtime/CL/functions/CLReshapeLayer.cpp
index 9abaa1b4e1..060eddb96c 100644
--- a/src/runtime/CL/functions/CLReshapeLayer.cpp
+++ b/src/runtime/CL/functions/CLReshapeLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -23,32 +23,21 @@
  */
 #include "arm_compute/runtime/CL/functions/CLReshapeLayer.h"
 
+#include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/ICLTensor.h"
-#include "src/core/CL/kernels/CLReshapeLayerKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/core/Validate.h"
+#include "src/core/CL/ICLKernel.h"
+#include "src/runtime/gpu/cl/operators/ClReshape.h"
 
 /** [CLReshapeLayer snippet] **/
 namespace arm_compute
 {
-namespace experimental
-{
-void CLReshape::configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output)
-{
-    auto k = std::make_unique<CLReshapeLayerKernel>();
-    k->configure(compile_context, input, output);
-    _kernel = std::move(k);
-}
-
-Status CLReshape::validate(const ITensorInfo *input, const ITensorInfo *output)
-{
-    return arm_compute::CLReshapeLayerKernel::validate(input, output);
-}
-} // namespace experimental
-
 struct CLReshapeLayer::Impl
 {
-    const ICLTensor                         *src{ nullptr };
-    ICLTensor                               *dst{ nullptr };
-    std::unique_ptr<experimental::CLReshape> op{ nullptr };
+    const ICLTensor                   *src{ nullptr };
+    ICLTensor                         *dst{ nullptr };
+    std::unique_ptr<opencl::ClReshape> op{ nullptr };
 };
 
 CLReshapeLayer::CLReshapeLayer()
@@ -69,14 +58,14 @@ void CLReshapeLayer::configure(const CLCompileContext &compile_context, const IC
 {
     _impl->src = input;
     _impl->dst = output;
-    _impl->op  = std::make_unique<experimental::CLReshape>();
+    _impl->op  = std::make_unique<opencl::ClReshape>();
     _impl->op->configure(compile_context, input->info(), output->info());
 }
 
 Status CLReshapeLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(experimental::CLReshape::validate(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(opencl::ClReshape::validate(input, output));
 
     return Status{};
 }
@@ -89,4 +78,4 @@ void CLReshapeLayer::run()
     _impl->op->run(pack);
 }
 } // namespace arm_compute
-/** [CLReshapeLayer snippet] **/
+/** [CLReshapeLayer snippet] **/
\ No newline at end of file
diff --git a/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp b/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp
index 2db064af44..6180f4de07 100644
--- a/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp
+++ b/src/runtime/CL/functions/CLSpaceToBatchLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2020 Arm Limited.
+ * Copyright (c) 2018-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -29,14 +29,13 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/Validate.h"
 #include "arm_compute/runtime/CL/CLScheduler.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
 #include "src/core/CL/kernels/CLSpaceToBatchLayerKernel.h"
 
 namespace arm_compute
 {
 CLSpaceToBatchLayer::CLSpaceToBatchLayer()
     : _space_to_batch_kernel(std::make_unique<CLSpaceToBatchLayerKernel>()),
-      _memset_kernel(std::make_unique<CLMemsetKernel>()),
+      _fill(),
       _has_padding(false)
 {
 }
@@ -55,7 +54,7 @@ void CLSpaceToBatchLayer::configure(const CLCompileContext &compile_context, con
     if(input->info()->tensor_shape().total_size() != output->info()->tensor_shape().total_size())
     {
         _has_padding = true;
-        _memset_kernel->configure(compile_context, output, PixelValue(0, input->info()->data_type(), input->info()->quantization_info()));
+        _fill.configure(compile_context, output, PixelValue(0, input->info()->data_type(), input->info()->quantization_info()));
     }
     _space_to_batch_kernel->configure(compile_context, input, block_shape, paddings, output);
 }
@@ -73,14 +72,14 @@ void CLSpaceToBatchLayer::configure(const CLCompileContext &compile_context, con
     if(input->info()->tensor_shape().total_size() != output->info()->tensor_shape().total_size())
     {
         _has_padding = true;
-        _memset_kernel->configure(compile_context, output, PixelValue(0, input->info()->data_type(), input->info()->quantization_info()));
+        _fill.configure(compile_context, output, PixelValue(0, input->info()->data_type(), input->info()->quantization_info()));
     }
     _space_to_batch_kernel->configure(compile_context, input, block_shape_x, block_shape_y, padding_left, padding_right, output);
 }
 
 Status CLSpaceToBatchLayer::validate(const ITensorInfo *input, const ITensorInfo *block_shape, const ITensorInfo *paddings, const ITensorInfo *output)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(CLMemsetKernel::validate(output, PixelValue(0, input->data_type(), input->quantization_info())));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLFill::validate(output, PixelValue(0, input->data_type(), input->quantization_info())));
     ARM_COMPUTE_RETURN_ON_ERROR(CLSpaceToBatchLayerKernel::validate(input, block_shape, paddings, output));
 
     return Status{};
@@ -89,7 +88,7 @@ Status CLSpaceToBatchLayer::validate(const ITensorInfo *input, const ITensorInfo
 Status CLSpaceToBatchLayer::validate(const ITensorInfo *input, const int block_shape_x, const int block_shape_y, const Size2D &padding_left, const Size2D &padding_right,
                                      const ITensorInfo *output)
 {
-    ARM_COMPUTE_RETURN_ON_ERROR(CLMemsetKernel::validate(output, PixelValue(0, input->data_type(), input->quantization_info())));
+    ARM_COMPUTE_RETURN_ON_ERROR(CLFill::validate(output, PixelValue(0, input->data_type(), input->quantization_info())));
     ARM_COMPUTE_RETURN_ON_ERROR(CLSpaceToBatchLayerKernel::validate(input, block_shape_x, block_shape_y, padding_left, padding_right, output));
 
     return Status{};
@@ -100,7 +99,8 @@ void CLSpaceToBatchLayer::run()
     // Zero out output only if we have paddings
     if(_has_padding)
     {
-        CLScheduler::get().enqueue(*_memset_kernel, true);
+        //CLScheduler::get().enqueue(*_fill, true);
+        _fill.run();
     }
     CLScheduler::get().enqueue(*_space_to_batch_kernel, true);
 }
diff --git a/src/runtime/gpu/cl/operators/ClCopy.cpp b/src/runtime/gpu/cl/operators/ClCopy.cpp
new file mode 100644
index 0000000000..2bdb1f5ba1
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClCopy.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClCopy.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClCopyKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClCopy::configure(const ClCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, Window *dst_window)
+{
+    auto k = std::make_unique<kernels::ClCopyKernel>();
+    k->configure(compile_context, src, dst, dst_window);
+    _kernel = std::move(k);
+}
+
+Status ClCopy::validate(const ITensorInfo *src, const ITensorInfo *dst, Window *dst_window)
+{
+    return kernels::ClCopyKernel::validate(src, dst, dst_window);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClCopy.h b/src/runtime/gpu/cl/operators/ClCopy.h
new file mode 100644
index 0000000000..0b99676f65
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClCopy.h
@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_COPY_H
+#define ARM_COMPUTE_CL_COPY_H
+
+#include "arm_compute/core/Window.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run @ref kernels::ClCopyKernel */
+class ClCopy : public IClOperator
+{
+public:
+    /** Constructor */
+    ClCopy() = default;
+    /** Initialise the function's source and destination.
+     *
+     * @param[in]  compile_context The compile context to be used.
+     * @param[in]  src             Source tensor info. Data types supported: All.
+     * @param[out] dst             Output tensor info. Data types supported: Same as @p src.
+     * @param[in]  dst_window      (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
+     *
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *src, ITensorInfo *dst, Window *dst_window = nullptr);
+    /** Static function to check if given info will lead to a valid configuration of @ref kernels::ClCopyKernel
+     *
+     * @param[in] src        Source tensor info. Data types supported: All.
+     * @param[in] dst        Output tensor info. Data types supported: Same as @p src.
+     * @param[in] dst_window (Optional) Window to be used in case only copying into part of a tensor. Default is nullptr.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, Window *dst_window = nullptr);
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_COPY_H */
diff --git a/src/runtime/gpu/cl/operators/ClCrop.cpp b/src/runtime/gpu/cl/operators/ClCrop.cpp
new file mode 100644
index 0000000000..17bb11912f
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClCrop.cpp
@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClCrop.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClCropKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClCrop::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value,
+                       Window *dst_window)
+{
+    auto k = std::make_unique<kernels::ClCropKernel>();
+    k->configure(compile_context, src, dst, start, end, batch_index, extrapolation_value, dst_window);
+    _kernel = std::move(k);
+}
+
+Status ClCrop::validate(const ITensorInfo *src, const ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value, Window *dst_window)
+{
+    return kernels::ClCropKernel::validate(src, dst, start, end, batch_index, extrapolation_value, dst_window);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClCrop.h b/src/runtime/gpu/cl/operators/ClCrop.h
new file mode 100644
index 0000000000..acfbf14742
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClCrop.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_COPY_H
+#define ARM_COMPUTE_CL_COPY_H
+
+#include "arm_compute/core/Window.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run @ref kernels::ClCropKernel */
+class ClCrop : public IClOperator
+{
+public:
+    /** Constructor */
+    ClCrop() = default;
+    /** Initialise the function's source and destination.
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in]  compile_context     The compile context to be used.
+     * @param[in]  src                 Source tensor info. Data type supported: All. Data layouts supported: NHWC.
+     * @param[out] dst                 Destination tensor info. Data type supported: F32
+     * @param[in]  start               Coordinates of where to start cropping the image.
+     * @param[in]  end                 Coordinates of where to end cropping the image.
+     * @param[in]  batch_index         Fourth dimension index of the 3D image to crop in @p src.
+     * @param[in]  extrapolation_value Value to be used for values outside of the image. Default is 0.
+     * @param[in]  dst_window          Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0,
+                   Window *dst_window = nullptr);
+
+    /** Static function to check if given info will lead to a valid configuration of @ref kernels::ClCropKernel
+     *
+     * @note Supported tensor rank: up to 4
+     *
+     * @param[in] src                 Source tensor info. Data type supported: All. Data layouts supported: NHWC.
+     * @param[in] dst                 Destination tensor info. Data type supported: F32
+     * @param[in] start               Coordinates of where to start cropping the image.
+     * @param[in] end                 Coordinates of where to end cropping the image.
+     * @param[in] batch_index         Fourth dimension index of the 3D image to crop in @p src.
+     * @param[in] extrapolation_value Value to be used for values outside of the image. Default is 0.
+     * @param[in] dst_window          Output window to be used in case cropped image is being copied into a tensor. Default is nullptr.
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, Coordinates2D start, Coordinates2D end, uint32_t batch_index, float extrapolation_value = 0,
+                           Window *dst_window = nullptr);
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_COPY_H */
diff --git a/src/runtime/gpu/cl/operators/ClFill.cpp b/src/runtime/gpu/cl/operators/ClFill.cpp
new file mode 100644
index 0000000000..4d0afaef24
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClFill.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClFill.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClFillKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClFill::configure(const ClCompileContext &compile_context, ITensorInfo *tensor, const PixelValue &constant_value, Window *dst_window)
+{
+    auto k = std::make_unique<kernels::ClFillKernel>();
+    k->configure(compile_context, tensor, constant_value, dst_window);
+    _kernel = std::move(k);
+}
+
+Status ClFill::validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *dst_window)
+{
+    return kernels::ClFillKernel::validate(tensor, constant_value, dst_window);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClFill.h b/src/runtime/gpu/cl/operators/ClFill.h
new file mode 100644
index 0000000000..e632d88546
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClFill.h
@@ -0,0 +1,61 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_FILL_H
+#define ARM_COMPUTE_CL_FILL_H
+
+#include "arm_compute/core/Window.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run @ref kernels::ClFillKernel */
+class ClFill : public IClOperator
+{
+public:
+    /** Constructor */
+    ClFill() = default;
+    /** Initialise the kernel's tensor and filling value
+     *
+     * @param[in]     compile_context The compile context to be used.
+     * @param[in,out] tensor          Source tensor info. Supported data types: All.
+     * @param[in]     constant_value  The value used to fill the planes of the tensor
+     * @param[in]     window          Window to be used in case setting only part of a tensor. Default is nullptr.
+     */
+    void configure(const CLCompileContext &compile_context, ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
+    /** Static function to check if given info will lead to a valid configuration of @ref kernels::ClFillKernel
+     *
+     * @param[in] tensor         Source tensor info. Data types supported: All.
+     * @param[in] constant_value The value used to fill the planes of the tensor.
+     * @param[in] window         Window to be used in case setting only part of a tensor. Default is nullptr.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *tensor, const PixelValue &constant_value, Window *window = nullptr);
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_FILL_H */
diff --git a/src/runtime/gpu/cl/operators/ClPermute.cpp b/src/runtime/gpu/cl/operators/ClPermute.cpp
new file mode 100644
index 0000000000..719bb6dac6
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClPermute.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClPermute.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClPermuteKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClPermute::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm)
+{
+    auto k = std::make_unique<kernels::ClPermuteKernel>();
+    k->configure(compile_context, src, dst, perm);
+    _kernel = std::move(k);
+}
+
+Status ClPermute::validate(const ITensorInfo *src, const ITensorInfo *dst, const PermutationVector &perm)
+{
+    return kernels::ClPermuteKernel::validate(src, dst, perm);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClPermute.h b/src/runtime/gpu/cl/operators/ClPermute.h
new file mode 100644
index 0000000000..65648c1434
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClPermute.h
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ARM_COMPUTE_CL_RESHAPE_H
+#define ARM_COMPUTE_CL_RESHAPE_H
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+/** Basic function to run @ref kernels::ClPermuteKernel */
+class ClPermute : public IClOperator
+{
+public:
+    /** Constructor */
+    ClPermute() = default;
+    /** Initialise the kernel's inputs and outputs and permute vector
+     *
+     * @note Arbitrary permutation vectors are supported with rank not greater than 4
+     *
+     * @param[in] compile_context The compile context to be used.
+     * @param[in] src             The src tensor info. Data types supported: All.
+     * @param[in] dst             The dst tensor info. Data types supported: Same as @p src
+     * @param[in] perm            Permutation vector
+     */
+    void configure(const CLCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst, const PermutationVector &perm);
+    /**  Static function to check if given info will lead to a valid configuration of @ref kernels::ClPermuteKernel.
+     *
+     * @note Arbitrary permutation vectors are supported with rank not greater than 4
+     *
+     * @param[in] src  First tensor src info. Data types supported: All.
+     * @param[in] dst  Output tensor info. Data types supported: same as @p src.
+     * @param[in] perm Permutation vector
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *src, const ITensorInfo *dst, const PermutationVector &perm);
+};
+} // namespace opencl
+} // namespace arm_compute
+#endif /* ARM_COMPUTE_CL_RESHAPE_H */
\ No newline at end of file
diff --git a/src/runtime/gpu/cl/operators/ClReshape.cpp b/src/runtime/gpu/cl/operators/ClReshape.cpp
new file mode 100644
index 0000000000..d3fa9f10ab
--- /dev/null
+++ b/src/runtime/gpu/cl/operators/ClReshape.cpp
@@ -0,0 +1,45 @@
+/*
+ * Copyright (c) 2021 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/gpu/cl/operators/ClReshape.h"
+
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/core/gpu/cl/kernels/ClReshapeKernel.h"
+
+namespace arm_compute
+{
+namespace opencl
+{
+void ClReshape::configure(const ClCompileContext &compile_context, const ITensorInfo *src, ITensorInfo *dst)
+{
+    auto k = std::make_unique<kernels::ClReshapeKernel>();
+    k->configure(compile_context, src, dst);
+    _kernel = std::move(k);
+}
+
+Status ClReshape::validate(const ITensorInfo *src, const ITensorInfo *dst)
+{
+    return kernels::ClReshapeKernel::validate(src, dst);
+}
+} // namespace opencl
+} // namespace arm_compute
\ No newline at end of file
diff --git a/src/core/CL/kernels/CLReshapeLayerKernel.h b/src/runtime/gpu/cl/operators/ClReshape.h
index 902c44649b..8cccc5776c 100644
--- a/src/core/CL/kernels/CLReshapeLayerKernel.h
+++ b/src/runtime/gpu/cl/operators/ClReshape.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,39 +21,39 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_CLRESHAPELAYERKERNEL_H
-#define ARM_COMPUTE_CLRESHAPELAYERKERNEL_H
+#ifndef ARM_COMPUTE_CL_RESHAPE_H
+#define ARM_COMPUTE_CL_RESHAPE_H
 
-#include "arm_compute/core/Types.h"
-#include "src/core/CL/ICLKernel.h"
+#include "src/core/gpu/cl/ClCompileContext.h"
+#include "src/runtime/gpu/cl/IClOperator.h"
 
 namespace arm_compute
 {
-class ICLTensor;
-
-/** Interface for the kernel to perform tensor reshaping */
-class CLReshapeLayerKernel : public ICLKernel
+namespace opencl
+{
+/** Basic function to run @ref kernels::ClReshapeKernel */
+class ClReshape : public IClOperator
 {
 public:
-    /** Set the input and output of the kernel
+    /** Constructor */
+    ClReshape() = default;
+    /** Initialise the kernel's inputs and outputs
      *
      * @param[in]  compile_context The compile context to be used.
-     * @param[in]  input           Source tensor info. Data type supported: All.
-     * @param[out] output          Destination tensor info. Data type supported: Same as @p input
+     * @param[in]  input           Input tensor info. Data type supported: All
+     * @param[out] output          Output info. Data type supported: Same as @p input
      */
     void configure(const CLCompileContext &compile_context, const ITensorInfo *input, ITensorInfo *output);
 
-    /** Static function to check if given info will lead to a valid configuration of @ref CLReshapeLayerKernel
+    /** Static function to check if given info will lead to a valid configuration of @ref kernels::ClReshapeKernel
      *
-     * @param[in] input  Source tensor info. Data type supported: All
-     * @param[in] output Destination tensor info. Data type supported: Same as @p input
+     * @param[in] input  Input tensor info. Data type supported: All
+     * @param[in] output Output tensor info. Data type supported: Same as @p input
      *
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output);
-
-    // Inherited methods overridden:
-    void run_op(ITensorPack &tensors, const Window &window, cl::CommandQueue &queue) override;
 };
+} // namespace opencl
 } // namespace arm_compute
-#endif /*ARM_COMPUTE_CLRESHAPELAYERKERNEL_H */
+#endif /* ARM_COMPUTE_CL_RESHAPE_H */
\ No newline at end of file
diff --git a/tests/CL/Helper.h b/tests/CL/Helper.h
index d217af6e18..5153e98add 100644
--- a/tests/CL/Helper.h
+++ b/tests/CL/Helper.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,12 +24,11 @@
 #ifndef ARM_COMPUTE_TEST_CL_HELPER_H
 #define ARM_COMPUTE_TEST_CL_HELPER_H
 
-#include "src/core/CL/kernels/CLFillBorderKernel.h"
-#include "src/core/CL/kernels/CLMemsetKernel.h"
-
 #include "arm_compute/runtime/CL/CLScheduler.h"
 #include "arm_compute/runtime/CL/ICLSimpleFunction.h"
+#include "arm_compute/runtime/CL/functions/CLFill.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "src/core/CL/kernels/CLFillBorderKernel.h"
 
 #include "src/core/CL/ICLKernel.h"
 
@@ -118,7 +117,7 @@ public:
         k->configure(first, second, std::forward<Args>(args)...);
         _kernel = std::move(k);
         _border_handler.configure(first, BorderSize(bordersize), BorderMode::CONSTANT, PixelValue());
-        _memset_kernel.configure(second, PixelValue());
+        _fill.configure(second, PixelValue());
     }
 
     // Inherited method overridden:
@@ -126,13 +125,13 @@ public:
     {
         ARM_COMPUTE_ERROR_ON_MSG(!_kernel, "The CL kernel or function isn't configured");
 
-        CLScheduler::get().enqueue(_memset_kernel, false);
+        _fill.run();
         CLScheduler::get().enqueue(_border_handler, false);
         CLScheduler::get().enqueue(*_kernel);
     }
 
 private:
-    CLMemsetKernel             _memset_kernel{};  /**< Kernel to initialize the tensor */
+    CLFill                     _fill{};           /**< Kernel to initialize the tensor */
     CLFillBorderKernel         _border_handler{}; /**< Kernel to handle  borders */
     std::unique_ptr<ICLKernel> _kernel{};         /**< Kernel to run */
 };