Improving Efficiency Of Deep Learning Models