Scalable Convolutional Neural Network Inference on-the-Edge using Low-Density FPGAs