Wireless Body Area Networks (WBANs) which are a subset of Edge Computing have been an integral part of Embedded and Wearable Systems. Up until now, there have been numerous efforts in domains like Computer Vision and Automatic Speech Recognition where several compression techniques have been applied to deep networks for facilitating low footprint deployment while maintaining expected efficiency. However, research and experimentations dealing with signal modality have been sparse particularly in the domains of healthcare. To reduce this gap, we propose the application of techniques with low computational overhead on our own baseline architectures primarily made of Convolutional Blocks and Classification Heads. We employ Depthwise Separable Convolutions and Low-Rank Tucker Decomposition on Vanilla Convolutional Neural Networks to reduce the number of learnable parameters. We successfully study the variations of compression paradigms in two directions, reduction in number of parameters and performance delta which provides a clear observation in trends seen for both, across increasing network depths. We obtain an average of 69.85% and 52.53% parameter reduction through Depthwise Separable and Tucker Decomposed Convolutions respectively, with minimal change in average performance (8.84% through Separable Convolutions & 5.59% through Low Rank Decomposition) on the validation partition set.