- brush up the implementation

 - add more channels/depth performance test for blend

"declare.in(img, WARMUP_RNG)" overrides input image with random values

Also added support for different input sizes (via cv::resize)

Added OCL performance test for YUV2RGB_IYUV, YUV2GRAY_420, RGB2YUV_IYUV, YUV2RGB_YUY2, YUV2GRAY_YUY2 color conversion modes

Fix kernel compilation errors on AMD system

Fix licanse information in cl file

Support CV_64F destination type

Change build options of the kernel

Optimize sum of square

Remove separate kernel for integral square

Increase epsilon for perfomance tests

Increase epsilon for perfomance tests

Test double support on AMD devices

Fix some issues

Try to fix problems with AMD device

Try to solve problem with AMD device

Fix error of destination size in kernel

Fix warnings