COMPMID-3375: Port NEActivationLayer functions/kernels to run on

different tensors.

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I98782bb73e9dc0899ffb1796aca6f99714adea94
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3343
Reviewed-by: Michalis Spyrou <michalis.spyrou@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

26 files changed, 281 insertions, 229 deletions

diff --git a/arm_compute/core/CPP/ICPPKernel.h b/arm_compute/core/CPP/ICPPKernel.h
index 21f6ab714a..3ec54756a0 100644
--- a/arm_compute/core/CPP/ICPPKernel.h
+++ b/arm_compute/core/CPP/ICPPKernel.h
@@ -84,7 +84,7 @@ public:
      * @param[in] window  Region on which to execute the kernel. (Must be a region of the window returned by window())
      * @param[in] info    Info about executing thread and CPU.
      */
-    virtual void run_op(const std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, const Window &window, const ThreadInfo &info)
+    virtual void run_op(const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs, const Window &window, const ThreadInfo &info)
     {
         ARM_COMPUTE_UNUSED(inputs, outputs, window, info);
     }
diff --git a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
index 82103b988b..399afa63c6 100644
--- a/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEActivationLayerKernel.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -33,6 +33,7 @@
 
 namespace arm_compute
 {
+// Forward declarations
 class ITensor;
 
 /** Interface for the activation layer kernel. */
@@ -57,12 +58,12 @@ public:
      *
      * @note If the output tensor is a nullptr, the activation function will be performed in-place
      *
-     * @param[in, out] input           Source tensor. In case of @p output tensor = nullptr, this tensor will store the result
+     * @param[in, out] input           Source tensor info. In case of @p output tensor = nullptr, this tensor will store the result
      *                                 of the activation function. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
-     * @param[out]     output          Destination tensor. Data type supported: same as @p input
+     * @param[out]     output          Destination tensor info. Data type supported: same as @p input
      * @param[in]      activation_info Activation layer information.
      */
-    void configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info);
+    void configure(const ITensorInfo *input, ITensorInfo *output, ActivationLayerInfo activation_info);
     /** Static function to check if given info will lead to a valid configuration of @ref NEActivationLayerKernel
      *
      * @param[in] input    Source tensor info. In case of @p output tensor info = nullptr, this tensor will store the result
@@ -75,7 +76,8 @@ public:
     static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info);
 
     // Inherited methods overridden:
-    void run(const Window &window, const ThreadInfo &info) override;
+    void run_op(const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs,
+                const Window &window, const ThreadInfo &info) override;
 
 private:
     using ActivationFunction = ActivationLayerInfo::ActivationFunction;
@@ -83,36 +85,34 @@ private:
      *
      * @param[in] window Region on which to execute the kernel.
      */
-    using ActivationFunctionExecutorPtr = void (NEActivationLayerKernel::*)(const Window &window);
+    using ActivationFunctionExecutorPtr = void (NEActivationLayerKernel::*)(const ITensor *src, ITensor *dst, const Window &window);
     /** Function to apply an activation function on a tensor.
      *
      * @param[in] window Region on which to execute the kernel
      */
     template <ActivationLayerInfo::ActivationFunction F, typename T>
     typename std::enable_if<arm_compute::utils::traits::is_floating_point<T>::value, void>::type
-    activation(const Window &window);
+    activation(const ITensor *src, ITensor *dst, const Window &window);
     /** Function to apply an activation function on a tensor.
      *
      * @param[in] window Region on which to execute the kernel
      */
     template <ActivationLayerInfo::ActivationFunction F, typename T>
-    typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type activation(const Window &window);
+    typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type activation(const ITensor *src, ITensor *dst, const Window &window);
     /** Function to apply an activation function on a tensor.
      *
      * @param[in] window Region on which to execute the kernel
      */
     template <ActivationLayerInfo::ActivationFunction F, typename T>
-    typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type activation(const Window &window);
+    typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type activation(const ITensor *src, ITensor *dst, const Window &window);
     /** Function to apply an activation function on a tensor.
      *
      * @param[in] window Region on which to execute the kernel
      */
     template <ActivationLayerInfo::ActivationFunction F, typename T>
-    typename std::enable_if<std::is_same<T, qsymm16_t>::value, void>::type activation(const Window &window);
+    typename std::enable_if<std::is_same<T, qsymm16_t>::value, void>::type activation(const ITensor *src, ITensor *dst, const Window &window);
 
 private:
-    ITensor                      *_input;
-    ITensor                      *_output;
     ActivationFunctionExecutorPtr _func;
     ActivationLayerInfo           _act_info;
 };
diff --git a/arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h b/arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h
index 6f888e0914..7a4dce128d 100644
--- a/arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h
+++ b/arm_compute/core/NEON/kernels/NEReshapeLayerKernel.h
@@ -57,8 +57,8 @@ public:
     static Status validate(const ITensorInfo *input, const ITensorInfo *output);
 
     // Inherited methods overridden:
-    void run_op(const std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, const Window &window, const ThreadInfo &info) override;
+    void run_op(const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs,
+                const Window &window, const ThreadInfo &info) override;
 };
-
 } // namespace arm_compute
 #endif /*ARM_COMPUTE_NERESHAPELAYERKERNEL_H */
diff --git a/arm_compute/core/experimental/Types.h b/arm_compute/core/experimental/Types.h
index 6043db9ff4..2b5591872a 100644
--- a/arm_compute/core/experimental/Types.h
+++ b/arm_compute/core/experimental/Types.h
@@ -36,21 +36,43 @@ class ITensor;
 /** Memory type */
 enum class TensorType
 {
-    ACL_SRC   = 0,
-    ACL_SRC_0 = 0,
-    ACL_SRC_1 = 1,
-    ACL_SRC_2 = 2,
-    ACL_DST   = 30,
-    ACL_DST_0 = 30,
-    ACL_DST_1 = 31,
-    ACL_INT   = 50,
-    ACL_INT_0 = 50,
-    ACL_INT_1 = 51,
-    ACL_INT_2 = 52
+    ACL_UNKNOWN = -1,
+    ACL_SRC     = 0,
+    ACL_SRC_0   = 0,
+    ACL_SRC_1   = 1,
+    ACL_SRC_2   = 2,
+    ACL_DST     = 30,
+    ACL_DST_0   = 30,
+    ACL_DST_1   = 31,
+    ACL_INT     = 50,
+    ACL_INT_0   = 50,
+    ACL_INT_1   = 51,
+    ACL_INT_2   = 52
 };
-using InputOperatorTensors  = std::pair<TensorType /* tensor type */, const ITensor * /* tensor object */>;
-using OutputOperatorTensors = std::pair<TensorType /* tensor type */, ITensor * /* tensor object */>;
-using OperatorTensors       = OutputOperatorTensors;
+
+/** Input tensor aggregate */
+struct InputTensor
+{
+    InputTensor(TensorType type, const ITensor *tensor)
+        : type(type), tensor(tensor)
+    {
+    }
+
+    TensorType     type{ TensorType::ACL_UNKNOWN };
+    const ITensor *tensor{ nullptr };
+};
+/** Output tensor aggregate */
+struct OutputTensor
+{
+    OutputTensor(TensorType type, ITensor *tensor)
+        : type(type), tensor(tensor)
+    {
+    }
+
+    TensorType type{ TensorType::ACL_UNKNOWN };
+    ITensor   *tensor{ nullptr };
+};
+using OperatorTensor = OutputTensor;
 
 namespace experimental
 {
diff --git a/arm_compute/runtime/CPP/CPPScheduler.h b/arm_compute/runtime/CPP/CPPScheduler.h
index 78ad43c2b4..2ccb094fdb 100644
--- a/arm_compute/runtime/CPP/CPPScheduler.h
+++ b/arm_compute/runtime/CPP/CPPScheduler.h
@@ -77,7 +77,7 @@ public:
      * @param[in] inputs  Vector that contains the input tensors.
      * @param[in] outputs Vector that contains the output tensors.
      */
-    void schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs) override;
+    void schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs) override;
 
 protected:
     /** Will run the workloads in parallel using num_threads
@@ -87,7 +87,7 @@ protected:
     void run_workloads(std::vector<Workload> &workloads) override;
 
 private:
-    void schedule_common(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs);
+    void schedule_common(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs);
     struct Impl;
     std::unique_ptr<Impl> _impl;
 };
diff --git a/arm_compute/runtime/IOperator.h b/arm_compute/runtime/IOperator.h
index 110c935702..cf3c8b05a1 100644
--- a/arm_compute/runtime/IOperator.h
+++ b/arm_compute/runtime/IOperator.h
@@ -46,7 +46,7 @@ public:
      * @param[in] workspace Vector that contains the workspace tensors.
      *
      */
-    virtual void run(std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, std::vector<OperatorTensors *> &workspace) = 0;
+    virtual void run(std::vector<InputTensor> inputs, std::vector<OutputTensor> outputs, std::vector<OperatorTensor> workspace) = 0;
     /** Prepare the function for executing
      *
      * Any one off pre-processing step required by the function is handled here
@@ -55,7 +55,7 @@ public:
      *
      * @note Prepare stage might not need all the function's buffers' backing memory to be available in order to execute
      */
-    virtual void prepare(std::vector<OperatorTensors *> constants) = 0;
+    virtual void prepare(std::vector<OperatorTensor> constants) = 0;
 
     /** Return the memory requirements required by the workspace
      */
diff --git a/arm_compute/runtime/IScheduler.h b/arm_compute/runtime/IScheduler.h
index 02d0cef086..40da86fd10 100644
--- a/arm_compute/runtime/IScheduler.h
+++ b/arm_compute/runtime/IScheduler.h
@@ -157,7 +157,7 @@ public:
      * @param[in] inputs  Vector containing the input tensors.
      * @param[in] outputs Vector containing the output tensors.
      */
-    virtual void schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs) = 0;
+    virtual void schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs) = 0;
 
     /** Execute all the passed workloads
      *
diff --git a/arm_compute/runtime/NEON/INEOperator.h b/arm_compute/runtime/NEON/INEOperator.h
index 4467e6d5ab..2f6e18048d 100644
--- a/arm_compute/runtime/NEON/INEOperator.h
+++ b/arm_compute/runtime/NEON/INEOperator.h
@@ -54,8 +54,8 @@ public:
     INEOperator &operator=(INEOperator &&) = default;
 
     // Inherited methods overridden:
-    void run(std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, std::vector<OperatorTensors *> &workspace) override final;
-    void prepare(std::vector<OperatorTensors *> constants) override final;
+    void run(std::vector<InputTensor> inputs, std::vector<OutputTensor> outputs, std::vector<OperatorTensor> workspace) override final;
+    void prepare(std::vector<OperatorTensor> constants) override final;
 
 protected:
     std::unique_ptr<INEKernel> _kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEActivationLayer.h b/arm_compute/runtime/NEON/functions/NEActivationLayer.h
index 95901dc2d8..8ac2dae911 100644
--- a/arm_compute/runtime/NEON/functions/NEActivationLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEActivationLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -27,6 +27,7 @@
 #include "arm_compute/runtime/NEON/INESimpleFunctionNoBorder.h"
 
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/INEOperator.h"
 
 namespace arm_compute
 {
@@ -37,7 +38,7 @@ class ITensor;
  *
  * @note The function simulates an activation layer with the specified activation function.
  */
-class NEActivationLayer : public INESimpleFunctionNoBorder
+class NEActivationLayer : public IFunction
 {
 public:
     /** Constructor
@@ -45,14 +46,16 @@ public:
      * @param[in] ctx Runtime context to be used by the function
      */
     NEActivationLayer(IRuntimeContext *ctx = nullptr);
+    /** Destructor */
+    ~NEActivationLayer();
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NEActivationLayer(const NEActivationLayer &) = delete;
     /** Default move constructor */
-    NEActivationLayer(NEActivationLayer &&) = default;
+    NEActivationLayer(NEActivationLayer &&);
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NEActivationLayer &operator=(const NEActivationLayer &) = delete;
     /** Default move assignment operator */
-    NEActivationLayer &operator=(NEActivationLayer &&) = default;
+    NEActivationLayer &operator=(NEActivationLayer &&);
     /** [NEActivationLayer snippet] **/
     /** Set the input and output tensor.
      *
@@ -75,6 +78,41 @@ public:
      * @return a status
      */
     static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info);
+
+    // Inherited methods overridden
+    void run() override;
+
+private:
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
+};
+
+namespace experimental
+{
+/** Basic function to run @ref NEActivationLayerKernel */
+class NEActivationLayer : public INEOperator
+{
+public:
+    /** Set the input and output tensor.
+     *
+     * @param[in]  input           Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
+     * @param[out] output          Destination tensor info. Data type supported: same as @p input
+     * @param[in]  activation_info Activation layer parameters.
+     */
+    void configure(const ITensorInfo *input, ITensorInfo *output, const ActivationLayerInfo &activation_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref NEActivationLayer
+     *
+     * @param[in] input    Source tensor info. Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM16/F16/F32.
+     * @param[in] output   Destination tensor info. Data type supported: same as @p input
+     * @param[in] act_info Activation layer information.
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info);
+
+    // Inherited methods overridden:
+    MemoryRequirements workspace() const override;
 };
+} // namespace experimental
 } // namespace arm_compute
 #endif /* ARM_COMPUTE_NEACTIVATIONLAYER_H */
diff --git a/arm_compute/runtime/NEON/functions/NELSTMLayer.h b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
index e85e87b88e..64845115b8 100644
--- a/arm_compute/runtime/NEON/functions/NELSTMLayer.h
+++ b/arm_compute/runtime/NEON/functions/NELSTMLayer.h
@@ -151,28 +151,28 @@ private:
     NEArithmeticAddition            _accum_input_gate1;
     NEArithmeticSubtractionKernel   _subtract_input_gate;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_input_gate;
-    NEActivationLayerKernel         _activation_input_gate;
+    NEActivationLayer               _activation_input_gate;
     NEFullyConnectedLayer           _fully_connected_forget_gate;
     NEArithmeticAddition            _accum_forget_gate1;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_forget_gate;
-    NEActivationLayerKernel         _activation_forget_gate;
+    NEActivationLayer               _activation_forget_gate;
     NEFullyConnectedLayer           _fully_connected_cell_state;
     NEGEMM                          _gemm_cell_state1;
     NETransposeKernel               _transpose_cell_state;
     NEArithmeticAdditionKernel      _accum_cell_state1;
     NEArithmeticAdditionKernel      _accum_cell_state2;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_cell_state1;
-    NEActivationLayerKernel         _activation_cell_state;
-    NEActivationLayerKernel         _cell_clip;
+    NEActivationLayer               _activation_cell_state;
+    NEActivationLayer               _cell_clip;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_cell_state2;
     NEFullyConnectedLayer           _fully_connected_output;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_output_state1;
     NEArithmeticAddition            _accum_output1;
-    NEActivationLayerKernel         _activation_output;
-    NEActivationLayerKernel         _activation_output_state;
+    NEActivationLayer               _activation_output;
+    NEActivationLayer               _activation_output_state;
     NEPixelWiseMultiplicationKernel _pixelwise_mul_output_state2;
     NEFullyConnectedLayer           _fully_connected_output_state;
-    NEActivationLayerKernel         _projection_clip;
+    NEActivationLayer               _projection_clip;
     NECopyKernel                    _copy_cell_state;
     NECopyKernel                    _copy_output;
     NEConcatenateLayer              _concat_scratch_buffer;
diff --git a/arm_compute/runtime/NEON/functions/NERNNLayer.h b/arm_compute/runtime/NEON/functions/NERNNLayer.h
index 0bfb905e19..db4134fd2d 100644
--- a/arm_compute/runtime/NEON/functions/NERNNLayer.h
+++ b/arm_compute/runtime/NEON/functions/NERNNLayer.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018-2019 ARM Limited.
+ * Copyright (c) 2018-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,11 +24,11 @@
 #ifndef ARM_COMPUTE_NERNNLAYER_H
 #define ARM_COMPUTE_NERNNLAYER_H
 
-#include "arm_compute/core/NEON/kernels/NEActivationLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEArithmeticAdditionKernel.h"
 #include "arm_compute/core/NEON/kernels/NECopyKernel.h"
 
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEFullyConnectedLayer.h"
 #include "arm_compute/runtime/NEON/functions/NEGEMM.h"
 
@@ -85,7 +85,7 @@ private:
     MemoryGroup                _memory_group;
     NEGEMM                     _gemm_state_f;
     NEArithmeticAdditionKernel _add_kernel;
-    NEActivationLayerKernel    _activation_kernel;
+    NEActivationLayer          _activation;
     NEFullyConnectedLayer      _fully_connected;
     NECopyKernel               _copy_kernel;
     Tensor                     _fully_connected_out;
diff --git a/arm_compute/runtime/NEON/functions/NEReshapeLayer.h b/arm_compute/runtime/NEON/functions/NEReshapeLayer.h
index 5a296a776d..43605584e0 100644
--- a/arm_compute/runtime/NEON/functions/NEReshapeLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEReshapeLayer.h
@@ -39,6 +39,18 @@ class ITensor;
 class NEReshapeLayer : public IFunction
 {
 public:
+    /** Default Constructor */
+    NEReshapeLayer();
+    /** Default Destructor */
+    ~NEReshapeLayer();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEReshapeLayer(const NEReshapeLayer &) = delete;
+    /** Default move constructor */
+    NEReshapeLayer(NEReshapeLayer &&);
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    NEReshapeLayer &operator=(const NEReshapeLayer &) = delete;
+    /** Default move assignment operator */
+    NEReshapeLayer &operator=(NEReshapeLayer &&);
     /** Initialise the kernel's inputs and outputs
      *
      * @param[in]  input  Input tensor. Data type supported: All
@@ -59,12 +71,8 @@ public:
     void run() override;
 
 private:
-    const ITensor *_input
-    {
-        nullptr
-    };
-    ITensor                              *_output{ nullptr };
-    std::unique_ptr<NEReshapeLayerKernel> _kernel{ nullptr };
+    struct Impl;
+    std::unique_ptr<Impl> _impl;
 };
 
 namespace experimental
diff --git a/arm_compute/runtime/OMP/OMPScheduler.h b/arm_compute/runtime/OMP/OMPScheduler.h
index 8ed1705a97..b7c186a838 100644
--- a/arm_compute/runtime/OMP/OMPScheduler.h
+++ b/arm_compute/runtime/OMP/OMPScheduler.h
@@ -66,7 +66,7 @@ public:
      * @param[in] inputs  Vector containing the input tensors.
      * @param[in] outputs Vector containing the output tensors.
      */
-    void schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs) override;
+    void schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs) override;
 
 protected:
     /** Execute all the passed workloads
diff --git a/arm_compute/runtime/SingleThreadScheduler.h b/arm_compute/runtime/SingleThreadScheduler.h
index 8094758249..8a69a5be15 100644
--- a/arm_compute/runtime/SingleThreadScheduler.h
+++ b/arm_compute/runtime/SingleThreadScheduler.h
@@ -57,7 +57,7 @@ public:
      * @param[in] inputs  Vector containing the input tensors.
      * @param[in] outputs Vector containing the output tensors.
      */
-    void schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs) override;
+    void schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs) override;
 
 protected:
     /** Will run the workloads sequentially and in order.
diff --git a/src/core/NEON/kernels/NEActivationLayerKernel.cpp b/src/core/NEON/kernels/NEActivationLayerKernel.cpp
index ffbfd710f9..2c00a76305 100644
--- a/src/core/NEON/kernels/NEActivationLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEActivationLayerKernel.cpp
@@ -95,7 +95,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c
     return Status{};
 }
 
-std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITensorInfo *output)
+std::pair<Status, Window> validate_and_configure_window(const ITensorInfo *input, ITensorInfo *output)
 {
     // Configure kernel window
     Window win = calculate_max_window(*input, Steps());
@@ -116,23 +116,15 @@ std::pair<Status, Window> validate_and_configure_window(ITensorInfo *input, ITen
 } // namespace
 
 NEActivationLayerKernel::NEActivationLayerKernel()
-    : _input(nullptr), _output(nullptr), _func(nullptr), _act_info()
+    : _func(nullptr), _act_info()
 {
 }
 
-void NEActivationLayerKernel::configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info)
+void NEActivationLayerKernel::configure(const ITensorInfo *input, ITensorInfo *output, ActivationLayerInfo activation_info)
 {
-    ARM_COMPUTE_ERROR_ON_NULLPTR(input);
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input, output);
 
-    _input    = input;
     _act_info = activation_info;
-    _output   = input;
-
-    // Out-of-place calculation
-    if(output != nullptr)
-    {
-        _output = output;
-    }
 
     // Disabled activation, thus no operation needed
     if(!activation_info.enabled())
@@ -140,7 +132,7 @@ void NEActivationLayerKernel::configure(ITensor *input, ITensor *output, Activat
         _func = nullptr;
     }
 
-    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr, activation_info));
+    ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input, output, activation_info));
 
     // Activation functions : FP32
     static std::map<ActivationFunction, ActivationFunctionExecutorPtr> act_map_f32 =
@@ -218,7 +210,7 @@ void NEActivationLayerKernel::configure(ITensor *input, ITensor *output, Activat
 
     };
 
-    switch(input->info()->data_type())
+    switch(input->data_type())
     {
         case DataType::QASYMM8_SIGNED:
             _func = act_map_qasymm8_signed[activation_info.activation()];
@@ -242,14 +234,14 @@ void NEActivationLayerKernel::configure(ITensor *input, ITensor *output, Activat
     }
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input->info(), (output != nullptr) ? output->info() : nullptr);
+    auto win_config = validate_and_configure_window(input, output);
     ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
     ICPPKernel::configure(win_config.second);
 }
 
 template <ActivationLayerInfo::ActivationFunction F, typename T>
 typename std::enable_if<arm_compute::utils::traits::is_floating_point<T>::value, void>::type
-NEActivationLayerKernel::activation(const Window &window)
+NEActivationLayerKernel::activation(const ITensor *src, ITensor *dst, const Window &window)
 {
     /** NEON vector tag type. */
     using ExactTagType = typename wrapper::traits::neon_bitvector_tag_t<T, wrapper::traits::BitWidth::W128>;
@@ -262,16 +254,16 @@ NEActivationLayerKernel::activation(const Window &window)
     Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
     win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
 
-    Iterator input(_input, win_collapsed);
-    Iterator output(_output, win_collapsed);
+    Iterator input(src, win_collapsed);
+    Iterator output(dst, win_collapsed);
 
     // A small delta added to the input to prevent NAN values caused by zeros in inputs to SQRT
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
-    const auto delta = wrapper::vdup_n(static_cast<T>(1e-7), ExactTagType{});
+    const auto delta = wrapper::vdup_n(static_cast<T>(1e-7), ExactTagType {});
 #else  /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-    const auto delta = wrapper::vdup_n(static_cast<T>(1e-24), ExactTagType{});
+    const auto delta = wrapper::vdup_n(static_cast<T>(1e-24), ExactTagType {});
 #endif /* __ARM_FEATURE_FP16_VECTOR_ARITHMETIC */
-    const auto const_1     = wrapper::vdup_n(static_cast<T>(1.f), ExactTagType{});
+    const auto const_1     = wrapper::vdup_n(static_cast<T>(1.f), ExactTagType {});
     const auto const_0     = wrapper::vdup_n(static_cast<T>(0.f), ExactTagType{});
     const auto const_6     = wrapper::vdup_n(static_cast<T>(6.f), ExactTagType{});
     const auto const_3     = wrapper::vdup_n(static_cast<T>(3.f), ExactTagType{});
@@ -402,7 +394,7 @@ NEActivationLayerKernel::activation(const Window &window)
 }
 
 template <ActivationLayerInfo::ActivationFunction F, typename T>
-typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type NEActivationLayerKernel::activation(const Window &window)
+typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type NEActivationLayerKernel::activation(const ITensor *src, ITensor *dst, const Window &window)
 {
     const int                window_step_x  = 16 / sizeof(T);
     const auto               window_start_x = static_cast<int>(window.x().start());
@@ -412,11 +404,11 @@ typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type NEActivat
     Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
     win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
 
-    Iterator input(_input, win_collapsed);
-    Iterator output(_output, win_collapsed);
+    Iterator input(src, win_collapsed);
+    Iterator output(dst, win_collapsed);
 
-    const UniformQuantizationInfo qi_in           = _input->info()->quantization_info().uniform();
-    const UniformQuantizationInfo qi_out          = _output->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_in           = src->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_out          = dst->info()->quantization_info().uniform();
     const qasymm8x16_t            va              = vdupq_n_u8(quantize_qasymm8(_act_info.a(), qi_in));
     const qasymm8x16_t            vb              = vdupq_n_u8(quantize_qasymm8(_act_info.b(), qi_in));
     const qasymm8_t               a               = quantize_qasymm8(_act_info.a(), qi_in);
@@ -579,7 +571,7 @@ typename std::enable_if<std::is_same<T, qasymm8_t>::value, void>::type NEActivat
 }
 
 template <ActivationLayerInfo::ActivationFunction F, typename T>
-typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type NEActivationLayerKernel::activation(const Window &window)
+typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type NEActivationLayerKernel::activation(const ITensor *src, ITensor *dst, const Window &window)
 {
     const int                window_step_x  = 16 / sizeof(T);
     const auto               window_start_x = static_cast<int>(window.x().start());
@@ -589,11 +581,11 @@ typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type NE
     Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
     win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
 
-    Iterator input(_input, win_collapsed);
-    Iterator output(_output, win_collapsed);
+    Iterator input(src, win_collapsed);
+    Iterator output(dst, win_collapsed);
 
-    const UniformQuantizationInfo qi_in           = _input->info()->quantization_info().uniform();
-    const UniformQuantizationInfo qi_out          = _output->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_in           = src->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_out          = dst->info()->quantization_info().uniform();
     const qasymm8x16_signed_t     va              = vdupq_n_s8(quantize_qasymm8_signed(_act_info.a(), qi_in));
     const qasymm8x16_signed_t     vb              = vdupq_n_s8(quantize_qasymm8_signed(_act_info.b(), qi_in));
     const qasymm8_signed_t        a               = quantize_qasymm8_signed(_act_info.a(), qi_in);
@@ -756,7 +748,7 @@ typename std::enable_if<std::is_same<T, qasymm8_signed_t>::value, void>::type NE
 }
 
 template <ActivationLayerInfo::ActivationFunction F, typename T>
-typename std::enable_if<std::is_same<T, qsymm16_t>::value, void>::type NEActivationLayerKernel::activation(const Window &window)
+typename std::enable_if<std::is_same<T, qsymm16_t>::value, void>::type NEActivationLayerKernel::activation(const ITensor *src, ITensor *dst, const Window &window)
 {
     const int                window_step_x  = 16 / sizeof(T);
     const auto               window_start_x = static_cast<int>(window.x().start());
@@ -766,11 +758,11 @@ typename std::enable_if<std::is_same<T, qsymm16_t>::value, void>::type NEActivat
     Window win_collapsed = window.collapse_if_possible(window, Window::DimZ);
     win_collapsed.set(Window::DimX, Window::Dimension(0, 1, 1));
 
-    Iterator input(_input, win_collapsed);
-    Iterator output(_output, win_collapsed);
+    Iterator input(src, win_collapsed);
+    Iterator output(dst, win_collapsed);
 
-    const UniformQuantizationInfo qi_in    = _input->info()->quantization_info().uniform();
-    const UniformQuantizationInfo qi_out   = _output->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_in    = src->info()->quantization_info().uniform();
+    const UniformQuantizationInfo qi_out   = dst->info()->quantization_info().uniform();
     const auto                    vconst_1 = vdupq_n_f32(1.f);
     const float32x4_t             va_f32   = vdupq_n_f32(_act_info.a());
     const float32x4_t             vb_f32   = vdupq_n_f32(_act_info.b());
@@ -863,7 +855,9 @@ Status NEActivationLayerKernel::validate(const ITensorInfo *input, const ITensor
     return Status{};
 }
 
-void NEActivationLayerKernel::run(const Window &window, const ThreadInfo &info)
+void NEActivationLayerKernel::run_op(const std::vector<InputTensor> &inputs,
+                                     const std::vector<OutputTensor> &outputs,
+                                     const Window &window, const ThreadInfo &info)
 {
     // Early exit on disabled activation
     if(!_act_info.enabled())
@@ -876,5 +870,7 @@ void NEActivationLayerKernel::run(const Window &window, const ThreadInfo &info)
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(IKernel::window(), window);
     ARM_COMPUTE_ERROR_ON(_func == nullptr);
 
-    (this->*_func)(window);
+    ARM_COMPUTE_ERROR_ON(inputs.empty() || outputs.empty());
+
+    (this->*_func)(inputs[0].tensor, outputs[0].tensor, window);
 }
diff --git a/src/core/NEON/kernels/NEReshapeLayerKernel.cpp b/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
index 600f8f9bf1..c141eecf75 100644
--- a/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
+++ b/src/core/NEON/kernels/NEReshapeLayerKernel.cpp
@@ -86,29 +86,32 @@ void NEReshapeLayerKernel::configure(const ITensorInfo *input, ITensorInfo *outp
     INEKernel::configure(win);
 }
 
-void NEReshapeLayerKernel::run_op(const std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, const Window &window, const ThreadInfo &info)
+void NEReshapeLayerKernel::run_op(const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs, const Window &window, const ThreadInfo &info)
 {
     ARM_COMPUTE_UNUSED(info);
     ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this);
     ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window);
 
-    switch(inputs[0]->second->info()->data_type())
+    const auto src = inputs[0].tensor;
+    auto       dst = outputs[0].tensor;
+
+    switch(src->info()->data_type())
     {
         case DataType::U8:
         case DataType::S8:
         case DataType::QASYMM8:
         case DataType::QASYMM8_SIGNED:
-            reshape_tensor<uint8_t>(window, inputs[0]->second, outputs[0]->second);
+            reshape_tensor<uint8_t>(window, src, dst);
             break;
         case DataType::U16:
         case DataType::S16:
         case DataType::F16:
-            reshape_tensor<uint16_t>(window, inputs[0]->second, outputs[0]->second);
+            reshape_tensor<uint16_t>(window, src, dst);
             break;
         case DataType::U32:
         case DataType::S32:
         case DataType::F32:
-            reshape_tensor<uint32_t>(window, inputs[0]->second, outputs[0]->second);
+            reshape_tensor<uint32_t>(window, src, dst);
             break;
         default:
             ARM_COMPUTE_ERROR("Unsupported data type!");
diff --git a/src/runtime/CPP/CPPScheduler.cpp b/src/runtime/CPP/CPPScheduler.cpp
index db551590ea..41e1a2d647 100644
--- a/src/runtime/CPP/CPPScheduler.cpp
+++ b/src/runtime/CPP/CPPScheduler.cpp
@@ -363,7 +363,7 @@ void CPPScheduler::run_workloads(std::vector<IScheduler::Workload> &workloads)
 }
 #endif /* DOXYGEN_SKIP_THIS */
 
-void CPPScheduler::schedule_common(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs)
+void CPPScheduler::schedule_common(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs)
 {
     ARM_COMPUTE_ERROR_ON_MSG(!kernel, "The child class didn't set the kernel");
 
@@ -473,15 +473,15 @@ void CPPScheduler::schedule_common(ICPPKernel *kernel, const Hints &hints, std::
     }
 }
 
-void CPPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs)
+void CPPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs)
 {
     schedule_common(kernel, hints, inputs, outputs);
 }
 
 void CPPScheduler::schedule(ICPPKernel *kernel, const Hints &hints)
 {
-    std::vector<InputOperatorTensors *>  inputs;
-    std::vector<OutputOperatorTensors *> outputs;
+    const std::vector<InputTensor> inputs;
+    std::vector<OutputTensor>      outputs;
     schedule_common(kernel, hints, inputs, outputs);
 }
 } // namespace arm_compute
diff --git a/src/runtime/CPP/SingleThreadScheduler.cpp b/src/runtime/CPP/SingleThreadScheduler.cpp
index 777f84bec8..8257628090 100644
--- a/src/runtime/CPP/SingleThreadScheduler.cpp
+++ b/src/runtime/CPP/SingleThreadScheduler.cpp
@@ -49,7 +49,7 @@ void SingleThreadScheduler::schedule(ICPPKernel *kernel, const Hints &hints)
     kernel->run(kernel->window(), info);
 }
 
-void SingleThreadScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs)
+void SingleThreadScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs)
 {
     ARM_COMPUTE_UNUSED(hints);
     ThreadInfo info;
diff --git a/src/runtime/NEON/INEOperator.cpp b/src/runtime/NEON/INEOperator.cpp
index c24d5c47f1..78790856ee 100644
--- a/src/runtime/NEON/INEOperator.cpp
+++ b/src/runtime/NEON/INEOperator.cpp
@@ -33,7 +33,7 @@ INEOperator::INEOperator(IRuntimeContext *ctx)
 {
 }
 
-void INEOperator::run(std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs, std::vector<OperatorTensors *> &workspace)
+void INEOperator::run(std::vector<InputTensor> inputs, std::vector<OutputTensor> outputs, std::vector<OperatorTensor> workspace)
 {
     ARM_COMPUTE_UNUSED(workspace);
 
@@ -45,7 +45,7 @@ void INEOperator::run(std::vector<InputOperatorTensors *> &inputs, std::vector<O
     NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, inputs, outputs);
 }
 
-void INEOperator::prepare(std::vector<OperatorTensors *> constants)
+void INEOperator::prepare(std::vector<OperatorTensor> constants)
 {
     ARM_COMPUTE_UNUSED(constants);
 }
diff --git a/src/runtime/NEON/functions/NEActivationLayer.cpp b/src/runtime/NEON/functions/NEActivationLayer.cpp
index e4d1125c79..889ff6b1f4 100644
--- a/src/runtime/NEON/functions/NEActivationLayer.cpp
+++ b/src/runtime/NEON/functions/NEActivationLayer.cpp
@@ -23,25 +23,76 @@
  */
 #include "arm_compute/runtime/NEON/functions/NEActivationLayer.h"
 
+#include "arm_compute/core/Error.h"
 #include "arm_compute/core/NEON/kernels/NEActivationLayerKernel.h"
+#include "arm_compute/core/experimental/Types.h"
 #include "arm_compute/runtime/IRuntimeContext.h"
+#include "arm_compute/runtime/Tensor.h"
 #include "support/MemorySupport.h"
 
 namespace arm_compute
 {
-NEActivationLayer::NEActivationLayer(IRuntimeContext *ctx) // NOLINT
-    : INESimpleFunctionNoBorder(ctx)
+namespace experimental
 {
-}
-void NEActivationLayer::configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info)
+void NEActivationLayer::configure(const ITensorInfo *input, ITensorInfo *output, const ActivationLayerInfo &activation_info)
 {
     auto k = arm_compute::support::cpp14::make_unique<NEActivationLayerKernel>();
     k->configure(input, output, activation_info);
     _kernel = std::move(k);
 }
 
+Status NEActivationLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &activation_info)
+{
+    return NEActivationLayerKernel::validate(input, output, activation_info);
+}
+
+MemoryRequirements NEActivationLayer::workspace() const
+{
+    return MemoryRequirements{};
+}
+} // namespace experimental
+
+struct NEActivationLayer::Impl
+{
+    const ITensor                                   *src{ nullptr };
+    ITensor                                         *dst{ nullptr };
+    IRuntimeContext                                 *ctx{ nullptr };
+    std::unique_ptr<experimental::NEActivationLayer> op{ nullptr };
+};
+
+NEActivationLayer::NEActivationLayer(IRuntimeContext *ctx)
+    : _impl(support::cpp14::make_unique<Impl>())
+{
+    _impl->ctx = ctx;
+}
+
+NEActivationLayer::NEActivationLayer(NEActivationLayer &&) = default;
+
+NEActivationLayer &NEActivationLayer::operator=(NEActivationLayer &&) = default;
+
+NEActivationLayer::~NEActivationLayer() = default;
+
+void NEActivationLayer::configure(ITensor *input, ITensor *output, ActivationLayerInfo activation_info)
+{
+    ARM_COMPUTE_ERROR_ON_NULLPTR(input);
+
+    _impl->src = input;
+    _impl->dst = output == nullptr ? input : output;
+
+    _impl->op = arm_compute::support::cpp14::make_unique<experimental::NEActivationLayer>();
+    _impl->op->configure(_impl->src->info(), _impl->dst->info(), activation_info);
+}
+
 Status NEActivationLayer::validate(const ITensorInfo *input, const ITensorInfo *output, const ActivationLayerInfo &act_info)
 {
-    return NEActivationLayerKernel::validate(input, output, act_info);
+    return experimental::NEActivationLayer::validate(input, output, act_info);
+}
+
+void NEActivationLayer::run()
+{
+    const InputTensor src{ TensorType::ACL_SRC, _impl->src };
+    OutputTensor      dst{ TensorType::ACL_DST, _impl->dst };
+
+    _impl->op->run({ src }, { dst }, {});
 }
 } // namespace arm_compute
diff --git a/src/runtime/NEON/functions/NELSTMLayer.cpp b/src/runtime/NEON/functions/NELSTMLayer.cpp
index f9d445fe71..0a111363e3 100644
--- a/src/runtime/NEON/functions/NELSTMLayer.cpp
+++ b/src/runtime/NEON/functions/NELSTMLayer.cpp
@@ -474,7 +474,7 @@ Status NELSTMLayer::validate(const ITensorInfo *input,
                                                                               RoundingPolicy::TO_ZERO));
         ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&forget_gate, forget_gate_bias, &forget_gate, ConvertPolicy::SATURATE));
     }
-    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&forget_gate, &forget_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&forget_gate, &forget_gate, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
 
     // Validate input gate
     if(!lstm_params.has_cifg_opt())
@@ -508,7 +508,7 @@ Status NELSTMLayer::validate(const ITensorInfo *input,
             ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&input_gate, lstm_params.input_layer_norm_weights(), &input_gate, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
             ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&input_gate, lstm_params.input_gate_bias(), &input_gate, ConvertPolicy::SATURATE));
         }
-        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&input_gate, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&input_gate, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
     }
     else
     {
@@ -526,14 +526,14 @@ Status NELSTMLayer::validate(const ITensorInfo *input,
                                                                               RoundingPolicy::TO_ZERO));
         ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&cell_state_tmp, cell_bias, &cell_state_tmp, ConvertPolicy::SATURATE));
     }
-    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, nullptr, activation_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&cell_state_tmp, nullptr, activation_info));
     ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &input_gate, &cell_state_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
     ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &forget_gate, &cell_state_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
     ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&cell_state_tmp, &cell_state_tmp, &cell_state_tmp, ConvertPolicy::SATURATE));
     if(cell_threshold != 0.f)
     {
-        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -cell_threshold,
-                                                                                                                    cell_threshold)));
+        ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&cell_state_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -cell_threshold,
+                                                                                                              cell_threshold)));
     }
 
     // Validate output gate tmp
@@ -559,18 +559,18 @@ Status NELSTMLayer::validate(const ITensorInfo *input,
                                                                               RoundingPolicy::TO_ZERO));
         ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAddition::validate(&output_gate_tmp, output_gate_bias, &output_gate_tmp, ConvertPolicy::SATURATE));
     }
-    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&output_gate_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&output_gate_tmp, nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)));
 
     // Validate output state
-    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&cell_state_tmp, &cell_state_tmp, activation_info));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&cell_state_tmp, &cell_state_tmp, activation_info));
     ARM_COMPUTE_RETURN_ON_ERROR(NEPixelWiseMultiplicationKernel::validate(&cell_state_tmp, &output_gate_tmp, &output_gate_tmp, 1, ConvertPolicy::SATURATE, RoundingPolicy::TO_ZERO));
     if(lstm_params.has_projection())
     {
         ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(&output_gate_tmp, lstm_params.projection_weights(), lstm_params.projection_bias(), output_state_out));
         if(projection_threshold != 0.f)
         {
-            ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(output_state_out, output_state_out,
-                                                                          ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -projection_threshold, projection_threshold)));
+            ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(output_state_out, output_state_out,
+                                                                    ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, -projection_threshold, projection_threshold)));
         }
     }
 
@@ -612,7 +612,7 @@ void NELSTMLayer::run()
         NEScheduler::get().schedule(&_pixelwise_mul_forget_gate_coeff, Window::DimY);
         NEScheduler::get().schedule(&_accum_forget_gate_bias, Window::DimY);
     }
-    NEScheduler::get().schedule(&_activation_forget_gate, Window::DimY);
+    _activation_forget_gate.run();
 
     if(_run_cifg_opt)
     {
@@ -642,7 +642,7 @@ void NELSTMLayer::run()
             NEScheduler::get().schedule(&_pixelwise_mul_input_gate_coeff, Window::DimY);
             NEScheduler::get().schedule(&_accum_input_gate_bias, Window::DimY);
         }
-        NEScheduler::get().schedule(&_activation_input_gate, Window::DimY);
+        _activation_input_gate.run();
     }
 
     _fully_connected_cell_state.run();
@@ -655,14 +655,14 @@ void NELSTMLayer::run()
         NEScheduler::get().schedule(&_pixelwise_mul_cell_gate_coeff, Window::DimY);
         NEScheduler::get().schedule(&_accum_cell_gate_bias, Window::DimY);
     }
-    NEScheduler::get().schedule(&_activation_cell_state, Window::DimY);
+    _activation_cell_state.run();
     NEScheduler::get().schedule(&_pixelwise_mul_cell_state1, Window::DimY);
     NEScheduler::get().schedule(&_pixelwise_mul_cell_state2, Window::DimY);
     NEScheduler::get().schedule(&_accum_cell_state2, Window::DimY);
 
     if(_perform_cell_clipping)
     {
-        NEScheduler::get().schedule(&_cell_clip, Window::DimY);
+        _cell_clip.run();
     }
 
     _fully_connected_output.run();
@@ -677,9 +677,9 @@ void NELSTMLayer::run()
         NEScheduler::get().schedule(&_pixelwise_mul_output_gate_coeff, Window::DimY);
         NEScheduler::get().schedule(&_accum_output_gate_bias, Window::DimY);
     }
-    NEScheduler::get().schedule(&_activation_output, Window::DimY);
+    _activation_output.run();
 
-    NEScheduler::get().schedule(&_activation_output_state, Window::DimY);
+    _activation_output_state.run();
     NEScheduler::get().schedule(&_pixelwise_mul_output_state2, Window::DimY);
 
     if(_has_projection_weights)
@@ -687,7 +687,7 @@ void NELSTMLayer::run()
         _fully_connected_output_state.run();
         if(_perform_projection_clipping)
         {
-            NEScheduler::get().schedule(&_projection_clip, Window::DimY);
+            _projection_clip.run();
         }
     }
 
diff --git a/src/runtime/NEON/functions/NERNNLayer.cpp b/src/runtime/NEON/functions/NERNNLayer.cpp
index 154b060c3d..4a15777be9 100644
--- a/src/runtime/NEON/functions/NERNNLayer.cpp
+++ b/src/runtime/NEON/functions/NERNNLayer.cpp
@@ -34,8 +34,8 @@
 namespace arm_compute
 {
 NERNNLayer::NERNNLayer(std::shared_ptr<IMemoryManager> memory_manager)
-    : _memory_group(std::move(memory_manager)), _gemm_state_f(), _add_kernel(), _activation_kernel(), _fully_connected(memory_manager), _copy_kernel(), _fully_connected_out(), _gemm_output(),
-      _add_output(), _is_prepared(false)
+    : _memory_group(std::move(memory_manager)), _gemm_state_f(), _add_kernel(), _activation(), _fully_connected(memory_manager), _copy_kernel(), _fully_connected_out(), _gemm_output(), _add_output(),
+      _is_prepared(false)
 {
 }
 
@@ -60,7 +60,7 @@ Status NERNNLayer::validate(const ITensorInfo *input, const ITensorInfo *weights
 
     ARM_COMPUTE_RETURN_ON_ERROR(NEFullyConnectedLayer::validate(input, weights, bias, &shape_info));
     ARM_COMPUTE_RETURN_ON_ERROR(NEArithmeticAdditionKernel::validate(&shape_info, &shape_info, &shape_info, ConvertPolicy::SATURATE));
-    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayerKernel::validate(&shape_info, &shape_info, info));
+    ARM_COMPUTE_RETURN_ON_ERROR(NEActivationLayer::validate(&shape_info, &shape_info, info));
 
     return Status{};
 }
@@ -95,7 +95,7 @@ void NERNNLayer::configure(const ITensor *input, const ITensor *weights, const I
     _fully_connected_out.allocator()->allocate();
     _gemm_output.allocator()->allocate();
 
-    _activation_kernel.configure(&_add_output, hidden_state, info);
+    _activation.configure(&_add_output, hidden_state, info);
     _add_output.allocator()->allocate();
 
     _copy_kernel.configure(hidden_state, output);
@@ -112,7 +112,7 @@ void NERNNLayer::run()
     _gemm_state_f.run();
 
     NEScheduler::get().schedule(&_add_kernel, Window::DimY);
-    NEScheduler::get().schedule(&_activation_kernel, Window::DimY);
+    _activation.run();
 
     // copy hidden out to output
     NEScheduler::get().schedule(&_copy_kernel, Window::DimY);
diff --git a/src/runtime/NEON/functions/NEReshapeLayer.cpp b/src/runtime/NEON/functions/NEReshapeLayer.cpp
index 680abef026..daf358e7db 100644
--- a/src/runtime/NEON/functions/NEReshapeLayer.cpp
+++ b/src/runtime/NEON/functions/NEReshapeLayer.cpp
@@ -44,7 +44,7 @@ void NEReshapeLayer::configure(const ITensorInfo *input, ITensorInfo *output)
 
 Status NEReshapeLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
-    return arm_compute::NEReshapeLayer::validate(input, output);
+    return arm_compute::NEReshapeLayerKernel::validate(input, output);
 }
 
 MemoryRequirements NEReshapeLayer::workspace() const
@@ -53,32 +53,44 @@ MemoryRequirements NEReshapeLayer::workspace() const
 }
 } // namespace experimental
 
-void NEReshapeLayer::configure(const ITensor *input, ITensor *output)
+struct NEReshapeLayer::Impl
 {
-    _input  = input;
-    _output = output;
+    const ITensor                                *src{ nullptr };
+    ITensor                                      *dst{ nullptr };
+    std::unique_ptr<experimental::NEReshapeLayer> op{ nullptr };
+};
 
-    auto k = arm_compute::support::cpp14::make_unique<NEReshapeLayerKernel>();
-    k->configure(input->info(), output->info());
-    _kernel = std::move(k);
+NEReshapeLayer::NEReshapeLayer()
+    : _impl(support::cpp14::make_unique<Impl>())
+{
+}
+
+NEReshapeLayer::NEReshapeLayer(NEReshapeLayer &&) = default;
+
+NEReshapeLayer &NEReshapeLayer::operator=(NEReshapeLayer &&) = default;
+
+NEReshapeLayer::~NEReshapeLayer() = default;
+
+void NEReshapeLayer::configure(const ITensor *input, ITensor *output)
+{
+    _impl->src = input;
+    _impl->dst = output;
+    _impl->op  = arm_compute::support::cpp14::make_unique<experimental::NEReshapeLayer>();
+    _impl->op->configure(input->info(), output->info());
 }
 
 Status NEReshapeLayer::validate(const ITensorInfo *input, const ITensorInfo *output)
 {
     ARM_COMPUTE_RETURN_ERROR_ON_NULLPTR(input, output);
-    ARM_COMPUTE_RETURN_ON_ERROR(NEReshapeLayerKernel::validate(input, output));
+    ARM_COMPUTE_RETURN_ON_ERROR(experimental::NEReshapeLayer::validate(input, output));
 
     return Status{};
 }
 
 void NEReshapeLayer::run()
 {
-    InputOperatorTensors  src_0 = std::make_pair(TensorType::ACL_SRC, _input);
-    OutputOperatorTensors dst_0 = std::make_pair(TensorType::ACL_DST, _output);
-
-    std::vector<InputOperatorTensors *>  inputs  = { &src_0 };
-    std::vector<OutputOperatorTensors *> outputs = { &dst_0 };
-
-    NEScheduler::get().schedule_op(_kernel.get(), Window::DimY, inputs, outputs);
+    const InputTensor src{ TensorType::ACL_SRC, _impl->src };
+    OutputTensor      dst{ TensorType::ACL_DST, _impl->dst };
+    _impl->op->run({ src }, { dst }, {});
 }
 } // namespace arm_compute
diff --git a/src/runtime/OMP/OMPScheduler.cpp b/src/runtime/OMP/OMPScheduler.cpp
index a1851f03c3..6d6b285019 100644
--- a/src/runtime/OMP/OMPScheduler.cpp
+++ b/src/runtime/OMP/OMPScheduler.cpp
@@ -83,7 +83,7 @@ void OMPScheduler::schedule(ICPPKernel *kernel, const Hints &hints)
     }
 }
 
-void OMPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs)
+void OMPScheduler::schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs)
 {
     ARM_COMPUTE_ERROR_ON_MSG(!kernel, "The child class didn't set the kernel");
     ARM_COMPUTE_ERROR_ON_MSG(hints.strategy() == StrategyHint::DYNAMIC,
diff --git a/tests/framework/instruments/SchedulerTimer.cpp b/tests/framework/instruments/SchedulerTimer.cpp
index 58e1b56904..8729179a46 100644
--- a/tests/framework/instruments/SchedulerTimer.cpp
+++ b/tests/framework/instruments/SchedulerTimer.cpp
@@ -86,7 +86,7 @@ public:
         _kernels.push_back(std::move(info));
     }
 
-    void schedule_op(ICPPKernel *kernel, const Hints &hints, std::vector<InputOperatorTensors *> &inputs, std::vector<OutputOperatorTensors *> &outputs) override
+    void schedule_op(ICPPKernel *kernel, const Hints &hints, const std::vector<InputTensor> &inputs, const std::vector<OutputTensor> &outputs) override
     {
         _timer.start();
         _real_scheduler.schedule_op(kernel, hints, inputs, outputs);
diff --git a/tests/validation/NEON/ReshapeOperator.cpp b/tests/validation/NEON/ReshapeOperator.cpp
deleted file mode 100644
index 82e9768a2c..0000000000
--- a/tests/validation/NEON/ReshapeOperator.cpp
+++ /dev/null
@@ -1,78 +0,0 @@
-/*
- * Copyright (c) 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.
- */
-#include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/NEON/functions/NEReshapeLayer.h"
-#include "arm_compute/runtime/OperatorTensor.h"
-#include "arm_compute/runtime/Tensor.h"
-#include "arm_compute/runtime/TensorAllocator.h"
-#include "tests/AssetsLibrary.h"
-#include "tests/Globals.h"
-#include "tests/NEON/Accessor.h"
-#include "tests/Utils.h"
-#include "tests/framework/Asserts.h"
-#include "tests/framework/Macros.h"
-#include "tests/framework/datasets/Datasets.h"
-#include "tests/validation/Validation.h"
-
-namespace arm_compute
-{
-namespace test
-{
-namespace validation
-{
-TEST_SUITE(NEON)
-TEST_SUITE(ReshapeOperator)
-
-TEST_CASE(Run, framework::DatasetMode::ALL)
-{
-    // Create tensors and info
-    TensorInfo src_info(TensorShape(27U, 11U, 3U), 1, DataType::F32);
-    TensorInfo dst_info(TensorShape(27U, 11U, 3U), 1, DataType::F32);
-    Tensor     src = create_tensor<Tensor>(TensorShape(27U, 11U, 3U), DataType::F32, 1);
-    Tensor     dst = create_tensor<Tensor>(TensorShape(27U, 11U, 3U), DataType::F32, 1);
-
-    // Create and configure function
-    experimental::NEReshapeLayer reshape_operator;
-    reshape_operator.configure(&src_info, &dst_info);
-
-    // Allocate tensors
-    src.allocator()->allocate();
-    dst.allocator()->allocate();
-
-    InputOperatorTensors  src_0 = std::make_pair(TensorType::ACL_SRC, &src);
-    OutputOperatorTensors dst_0 = std::make_pair(TensorType::ACL_DST, &dst);
-
-    std::vector<InputOperatorTensors *>  src_vec  = { &src_0 };
-    std::vector<OutputOperatorTensors *> dst_vec  = { &dst_0 };
-    std::vector<OperatorTensors *>       work_vec = {};
-
-    // Compute functions
-    reshape_operator.run(src_vec, dst_vec, work_vec);
-}
-
-TEST_SUITE_END() // ReshapeOperator
-TEST_SUITE_END() // NEON
-} // namespace validation
-} // namespace test
-} // namespace arm_compute