Compressed Learning of Deep Neural Networks for OpenCL-Capable Embedded Systems