TinyML has emerged as a transformative technology enabling the development of tiny yet powerful models for on-edge real-time sensory processing. This work investigates how optimization techniques can improve the efficiency of TinyML models for deployment on resource-constrained devices like the Arduino Nano 33 BLE Sense while maintaining acceptable accuracy. Optimization techniques such as pruning followed by a sparsity removal approach and quantization were applied using the TensorFlow framework. The results obtained demonstrate a considerable reduction in processing time of around 98% and RAM use of 65% on the Arduino Nano 33 BLE Sense with a slight loss in accuracy by 4%.