Compression of Deep Convolutional Neural Networks for Fast and Low Power Mobile Applications