Speed up Deep Learning Models with Minimal Effort