COMPMID-534: Add MemoryManager support in NEON functions

Adds support for:
-NECannyEdge
-NEConvolution
-NEDirectConvolution
-NEGEMM
-NEGEMMLowp
-NEGaussian5x5
-NEHOGDescriptor
-NEHOGGradient
-NEL2Normalize
-NELocallyConnectedLayer
-NENormalizationLayer
-NEScale
-NESobel5x5
-NESobel7x7

Change-Id: I68e05aa6054372fa873a882633a15fb97882c00d
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/87926
Reviewed-by: Pablo Tello <pablo.tello@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>

16 files changed, 97 insertions, 22 deletions

diff --git a/arm_compute/runtime/NEON/functions/NECannyEdge.h b/arm_compute/runtime/NEON/functions/NECannyEdge.h
index fbf2d90740..b7e0ffbcf1 100644
--- a/arm_compute/runtime/NEON/functions/NECannyEdge.h
+++ b/arm_compute/runtime/NEON/functions/NECannyEdge.h
@@ -28,6 +28,8 @@
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
@@ -55,7 +57,7 @@ public:
      *
      * Initialize Sobel kernel to nullptr.
      */
-    NECannyEdge();
+    NECannyEdge(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NECannyEdge(const NECannyEdge &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
@@ -80,6 +82,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                   _memory_group;        /**< Function's memory group */
     std::unique_ptr<IFunction>    _sobel;               /**< Pointer to Sobel kernel */
     std::unique_ptr<INEKernel>    _gradient;            /**< Gradient kernel */
     NEEdgeNonMaxSuppressionKernel _non_max_suppr;       /**< Non-Maxima suppression kernel */
diff --git a/arm_compute/runtime/NEON/functions/NEConvolution.h b/arm_compute/runtime/NEON/functions/NEConvolution.h
index 1704d9fa94..9c0a906651 100644
--- a/arm_compute/runtime/NEON/functions/NEConvolution.h
+++ b/arm_compute/runtime/NEON/functions/NEConvolution.h
@@ -28,10 +28,13 @@
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/INESimpleFunction.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
@@ -70,7 +73,7 @@ class NEConvolutionSquare : public IFunction
 {
 public:
     /** Default constructor */
-    NEConvolutionSquare();
+    NEConvolutionSquare(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialize the function's source, destination, conv and border_mode.
      *
      * @param[in,out] input                 Source tensor. Data type supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -86,6 +89,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                                   _memory_group;   /**< Function memory group */
     Tensor                                        _tmp;            /**< temporary buffer for output of horizontal pass */
     bool                                          _is_separable;   /**< true if the convolution can be separated */
     NESeparableConvolutionHorKernel<matrix_size>  _kernel_hor;     /**< kernel for horizontal pass of separated convolution */
diff --git a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
index 9d2775ada6..daaf18f297 100644
--- a/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
+++ b/arm_compute/runtime/NEON/functions/NEDirectConvolutionLayer.h
@@ -29,8 +29,12 @@
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 /** Function to run the direct convolution.
@@ -45,7 +49,7 @@ class NEDirectConvolutionLayer : public IFunction
 {
 public:
     /** Constructor */
-    NEDirectConvolutionLayer();
+    NEDirectConvolutionLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Set the input, weights, biases and output tensors.
       *
       * @note: DirectConvolution only works in the following configurations:
@@ -69,6 +73,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                                  _memory_group;
     NEDirectConvolutionLayerBiasAccumulateKernel _accumulate_bias_kernel;
     NEDirectConvolutionLayerKernel               _conv_kernel;
     NEFillBorderKernel                           _input_border_handler;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMM.h b/arm_compute/runtime/NEON/functions/NEGEMM.h
index 3c8d7cf9b7..b4b9e8be01 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMM.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMM.h
@@ -30,8 +30,12 @@
 #include "arm_compute/core/NEON/kernels/NEGEMMMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 /** Basic function to execute GEMM on NEON. This function calls the following NEON kernels:
@@ -46,7 +50,7 @@ class NEGEMM : public IFunction
 {
 public:
     /** Constructor */
-    NEGEMM();
+    NEGEMM(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the kernel's inputs, output
      *
      * @note GEMM: General Matrix Multiply - [alpha * A * B + beta * C].
@@ -65,6 +69,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                _memory_group;
     NEGEMMInterleave4x4Kernel  _interleave_kernel;
     NEGEMMTranspose1xWKernel   _transpose_kernel;
     NEGEMMMatrixMultiplyKernel _mm_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEGEMMLowp.h b/arm_compute/runtime/NEON/functions/NEGEMMLowp.h
index bfb1a494b8..0b0a7742f6 100644
--- a/arm_compute/runtime/NEON/functions/NEGEMMLowp.h
+++ b/arm_compute/runtime/NEON/functions/NEGEMMLowp.h
@@ -32,6 +32,10 @@
 #include "arm_compute/core/NEON/kernels/NEGEMMInterleave4x4Kernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMLowpMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEGEMMTranspose1xWKernel.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
+
+#include <memory>
 
 namespace arm_compute
 {
@@ -48,7 +52,7 @@ class NEGEMMLowp : public IFunction
 {
 public:
     /** Constructor */
-    NEGEMMLowp();
+    NEGEMMLowp(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the kernel's inputs, output
     *
     * @note GEMM_LOWP:  low precision GEMM kernel
@@ -75,6 +79,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                    _memory_group;
     NEGEMMInterleave4x4Kernel      _interleave_kernel;
     NEGEMMTranspose1xWKernel       _transpose_kernel;
     NEGEMMLowpMatrixMultiplyKernel _mm_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEGaussian5x5.h b/arm_compute/runtime/NEON/functions/NEGaussian5x5.h
index 699e42efb4..2aae3cb513 100644
--- a/arm_compute/runtime/NEON/functions/NEGaussian5x5.h
+++ b/arm_compute/runtime/NEON/functions/NEGaussian5x5.h
@@ -28,9 +28,12 @@
 #include "arm_compute/core/NEON/kernels/NEGaussian5x5Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
@@ -48,7 +51,7 @@ class NEGaussian5x5 : public IFunction
 public:
     /** Default constructor
      */
-    NEGaussian5x5();
+    NEGaussian5x5(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the function's input, output and border mode.
      *
      * @param[in, out] input                 Source tensor. Data type supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -62,6 +65,7 @@ public:
     void run() override;
 
 protected:
+    MemoryGroup             _memory_group;   /**< Function memory group */
     NEGaussian5x5HorKernel  _kernel_hor;     /**< kernel for horizontal pass */
     NEGaussian5x5VertKernel _kernel_vert;    /**< kernel for vertical pass */
     Tensor                  _tmp;            /**< temporary buffer for output of horizontal pass */
diff --git a/arm_compute/runtime/NEON/functions/NEHOGDescriptor.h b/arm_compute/runtime/NEON/functions/NEHOGDescriptor.h
index b7b4909060..30989568e1 100644
--- a/arm_compute/runtime/NEON/functions/NEHOGDescriptor.h
+++ b/arm_compute/runtime/NEON/functions/NEHOGDescriptor.h
@@ -26,9 +26,13 @@
 
 #include "arm_compute/core/NEON/kernels/NEHOGDescriptorKernel.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/functions/NEHOGGradient.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 class IHOG;
@@ -43,7 +47,7 @@ class NEHOGDescriptor : public IFunction
 {
 public:
     /** Default constructor */
-    NEHOGDescriptor();
+    NEHOGDescriptor(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the function's source, destination, HOG data-object and border mode
      *
      * @param[in, out] input                 Input tensor. Data type supported: U8
@@ -59,6 +63,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                   _memory_group;
     NEHOGGradient                 _gradient;
     NEHOGOrientationBinningKernel _orient_bin;
     NEHOGBlockNormalizationKernel _block_norm;
diff --git a/arm_compute/runtime/NEON/functions/NEHOGGradient.h b/arm_compute/runtime/NEON/functions/NEHOGGradient.h
index dd2d99adfe..7e268411e1 100644
--- a/arm_compute/runtime/NEON/functions/NEHOGGradient.h
+++ b/arm_compute/runtime/NEON/functions/NEHOGGradient.h
@@ -27,6 +27,8 @@
 #include "arm_compute/core/NEON/INEKernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/functions/NEDerivative.h"
 #include "arm_compute/runtime/Tensor.h"
 
@@ -46,7 +48,7 @@ class NEHOGGradient : public IFunction
 {
 public:
     /** Default constructor */
-    NEHOGGradient();
+    NEHOGGradient(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the function's source, destinations, phase type and border mode
      *
      * @param[in, out] input                 Input tensor. Data type supported: U8.
@@ -63,6 +65,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                _memory_group;
     NEDerivative               _derivative;
     std::unique_ptr<INEKernel> _mag_phase;
     Tensor                     _gx;
diff --git a/arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h b/arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h
index 2d07e6435f..0d268ca565 100644
--- a/arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h
+++ b/arm_compute/runtime/NEON/functions/NEHOGMultiDetection.h
@@ -29,10 +29,14 @@
 #include "arm_compute/core/IMultiHOG.h"
 #include "arm_compute/core/NEON/kernels/NEHOGDescriptorKernel.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/functions/NEHOGDetector.h"
 #include "arm_compute/runtime/NEON/functions/NEHOGGradient.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 /** Basic function to detect multiple objects (or the same object at different scales) on the same input image using HOG. This function calls the following NEON kernels:
@@ -53,7 +57,7 @@ class NEHOGMultiDetection : public IFunction
 {
 public:
     /** Default constructor */
-    NEHOGMultiDetection();
+    NEHOGMultiDetection(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NEHOGMultiDetection(const NEHOGMultiDetection &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
@@ -85,6 +89,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                                                   _memory_group;
     NEHOGGradient                                                 _gradient_kernel;
     std::unique_ptr<NEHOGOrientationBinningKernel[]>              _orient_bin_kernel;
     std::unique_ptr<NEHOGBlockNormalizationKernel[]>              _block_norm_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NEL2Normalize.h b/arm_compute/runtime/NEON/functions/NEL2Normalize.h
index 1297b99e79..95d5186c13 100644
--- a/arm_compute/runtime/NEON/functions/NEL2Normalize.h
+++ b/arm_compute/runtime/NEON/functions/NEL2Normalize.h
@@ -26,9 +26,13 @@
 
 #include "arm_compute/core/NEON/kernels/NEL2NormalizeKernel.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/functions/NEReductionOperation.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 class ITensor;
@@ -43,7 +47,7 @@ class NEL2Normalize : public IFunction
 {
 public:
     /** Constructor */
-    NEL2Normalize();
+    NEL2Normalize(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Set the input and output tensors.
      *
      * @param[in, out] input   Source tensor. Data types supported: F32. (Written to only for border_size != 0)
@@ -57,6 +61,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup          _memory_group;
     NEReductionOperation _reduce_func;
     NEL2NormalizeKernel  _normalize_kernel;
     Tensor               _sumsq;
diff --git a/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h b/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
index efb2ff6c8b..18d2a1dfb5 100644
--- a/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
+++ b/arm_compute/runtime/NEON/functions/NELocallyConnectedLayer.h
@@ -31,8 +31,12 @@
 #include "arm_compute/core/NEON/kernels/NELocallyConnectedMatrixMultiplyKernel.h"
 #include "arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h"
 #include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
+#include <memory>
+
 namespace arm_compute
 {
 class INETensor;
@@ -48,7 +52,7 @@ class NELocallyConnectedLayer : public IFunction
 {
 public:
     /** Default constructor */
-    NELocallyConnectedLayer();
+    NELocallyConnectedLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Set the input and output tensors.
      *
      * @param[in]  input     Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
@@ -66,6 +70,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                            _memory_group;
     NEIm2ColKernel                         _input_im2col_kernel;
     NEWeightsReshapeKernel                 _weights_reshape_kernel;
     NELocallyConnectedMatrixMultiplyKernel _mm_kernel;
diff --git a/arm_compute/runtime/NEON/functions/NENormalizationLayer.h b/arm_compute/runtime/NEON/functions/NENormalizationLayer.h
index 4cfea226f3..1c95c5bc4a 100644
--- a/arm_compute/runtime/NEON/functions/NENormalizationLayer.h
+++ b/arm_compute/runtime/NEON/functions/NENormalizationLayer.h
@@ -29,9 +29,12 @@
 #include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
 #include "arm_compute/core/NEON/kernels/NENormalizationLayerKernel.h"
 #include "arm_compute/core/NEON/kernels/NEPixelWiseMultiplicationKernel.h"
+#include "arm_compute/core/Types.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
-#include "arm_compute/core/Types.h"
+#include <memory>
 
 namespace arm_compute
 {
@@ -48,7 +51,7 @@ class NENormalizationLayer : public IFunction
 {
 public:
     /** Default constructor */
-    NENormalizationLayer();
+    NENormalizationLayer(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Set the input and output tensors.
      *
      * @param[in]  input     Source tensor. 3 lower dims represent a single input with dimensions [width, height, IFM],
@@ -62,6 +65,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                     _memory_group;    /**< Function memory group */
     NENormalizationLayerKernel      _norm_kernel;     /**< Normalization layer kernel */
     NEPixelWiseMultiplicationKernel _multiply_kernel; /**< Pixel multiplication kernel */
     NEFillBorderKernel              _border_handler;  /**< Kernel to handle  borders */
diff --git a/arm_compute/runtime/NEON/functions/NEOpticalFlow.h b/arm_compute/runtime/NEON/functions/NEOpticalFlow.h
index 0534551d19..5d1fbe3a22 100644
--- a/arm_compute/runtime/NEON/functions/NEOpticalFlow.h
+++ b/arm_compute/runtime/NEON/functions/NEOpticalFlow.h
@@ -29,6 +29,8 @@
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/Array.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/NEON/functions/NEScharr3x3.h"
 #include "arm_compute/runtime/Tensor.h"
 
@@ -51,7 +53,7 @@ class NEOpticalFlow : public IFunction
 {
 public:
     /** Constructor */
-    NEOpticalFlow();
+    NEOpticalFlow(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Prevent instances of this class from being copied (As this class contains pointers) */
     NEOpticalFlow(const NEOpticalFlow &) = delete;
     /** Prevent instances of this class from being copied (As this class contains pointers) */
@@ -80,6 +82,7 @@ public:
     void run() override;
 
 private:
+    MemoryGroup                          _memory_group;
     std::unique_ptr<NEScharr3x3[]>       _func_scharr;
     std::unique_ptr<NELKTrackerKernel[]> _kernel_tracker;
     std::unique_ptr<Tensor[]>            _scharr_gx;
diff --git a/arm_compute/runtime/NEON/functions/NEScale.h b/arm_compute/runtime/NEON/functions/NEScale.h
index e1da891dcf..91cda066e7 100644
--- a/arm_compute/runtime/NEON/functions/NEScale.h
+++ b/arm_compute/runtime/NEON/functions/NEScale.h
@@ -24,25 +24,30 @@
 #ifndef __ARM_COMPUTE_NESCALEIMAGE_H__
 #define __ARM_COMPUTE_NESCALEIMAGE_H__
 
+#include "arm_compute/core/NEON/kernels/NEFillBorderKernel.h"
+#include "arm_compute/core/NEON/kernels/NEScaleKernel.h"
 #include "arm_compute/core/Types.h"
-#include "arm_compute/runtime/NEON/INESimpleFunction.h"
+#include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
 class ITensor;
 
 /** Basic function to run @ref NEScaleKernel */
-class NEScale : public INESimpleFunction
+class NEScale : public IFunction
 {
 public:
     /** Constructor
      *
      * Initialize NEScale
      */
-    NEScale();
+    NEScale(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialize the function's source, destination, interpolation type and border_mode.
      *
      * @param[in, out] input                 Source tensor. Data type supported: U8. (Written to only for @p border_mode != UNDEFINED)
@@ -53,10 +58,16 @@ public:
      */
     void configure(ITensor *input, ITensor *output, InterpolationPolicy policy, BorderMode border_mode, uint8_t constant_border_value = 0);
 
+    // Inherited methods overridden:
+    void run() override;
+
 private:
-    Tensor _offsets; /**< Offset to access the element with NEAREST interpolation or the top-left element with BILINEAR interpolation in the input tensor */
-    Tensor _dx;      /**< Element's distance between the X real coordinate and the smallest X following integer */
-    Tensor _dy;      /**< Element's distance between the Y real coordinate and the smallest Y following integer */
+    MemoryGroup        _memory_group;   /**< Function memory group */
+    Tensor             _offsets;        /**< Offset to access the element with NEAREST interpolation or the top-left element with BILINEAR interpolation in the input tensor */
+    Tensor             _dx;             /**< Element's distance between the X real coordinate and the smallest X following integer */
+    Tensor             _dy;             /**< Element's distance between the Y real coordinate and the smallest Y following integer */
+    NEScaleKernel      _scale_kernel;   /**< Kernel to perform the scaling */
+    NEFillBorderKernel _border_handler; /**< kernel to handle tensor borders */
 };
 }
 #endif /*__ARM_COMPUTE_NESCALEIMAGE_H__ */
diff --git a/arm_compute/runtime/NEON/functions/NESobel5x5.h b/arm_compute/runtime/NEON/functions/NESobel5x5.h
index fc4d665a70..2b7cb70f15 100644
--- a/arm_compute/runtime/NEON/functions/NESobel5x5.h
+++ b/arm_compute/runtime/NEON/functions/NESobel5x5.h
@@ -28,9 +28,12 @@
 #include "arm_compute/core/NEON/kernels/NESobel5x5Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
@@ -47,7 +50,7 @@ class NESobel5x5 : public IFunction
 {
 public:
     /** Default constructor */
-    NESobel5x5();
+    NESobel5x5(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the function's source, destinations and border mode.
      *
      * @note At least one of output_x or output_y must be not NULL.
@@ -65,6 +68,7 @@ public:
     void run() override;
 
 protected:
+    MemoryGroup          _memory_group;   /**< Function memory group */
     NESobel5x5HorKernel  _sobel_hor;      /**< Sobel Horizontal 5x5 kernel */
     NESobel5x5VertKernel _sobel_vert;     /**< Sobel Vertical 5x5 kernel */
     Tensor               _tmp_x;          /**< Temporary buffer for Sobel X */
diff --git a/arm_compute/runtime/NEON/functions/NESobel7x7.h b/arm_compute/runtime/NEON/functions/NESobel7x7.h
index 06b7c80ad6..5f7bab7cfd 100644
--- a/arm_compute/runtime/NEON/functions/NESobel7x7.h
+++ b/arm_compute/runtime/NEON/functions/NESobel7x7.h
@@ -28,9 +28,12 @@
 #include "arm_compute/core/NEON/kernels/NESobel7x7Kernel.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/runtime/IFunction.h"
+#include "arm_compute/runtime/IMemoryManager.h"
+#include "arm_compute/runtime/MemoryGroup.h"
 #include "arm_compute/runtime/Tensor.h"
 
 #include <cstdint>
+#include <memory>
 
 namespace arm_compute
 {
@@ -47,7 +50,7 @@ class NESobel7x7 : public IFunction
 {
 public:
     /** Default constructor */
-    NESobel7x7();
+    NESobel7x7(std::shared_ptr<IMemoryManager> memory_manager = nullptr);
     /** Initialise the function's source, destinations and border mode.
      *
      * @note At least one of output_x or output_y must be not NULL.
@@ -65,6 +68,7 @@ public:
     void run() override;
 
 protected:
+    MemoryGroup          _memory_group;   /**< Function memory group */
     NESobel7x7HorKernel  _sobel_hor;      /**< Sobel Horizontal 7x7 kernel */
     NESobel7x7VertKernel _sobel_vert;     /**< Sobel Vertical 7x7 kernel */
     Tensor               _tmp_x;          /**< Temporary buffer for Sobel X */