Exploring techniques to build efficient and robust TinyML deployments