As the Internet of Things based wearable medical devices become more widely adopted for supporting early diagnosis and remote health monitoring, the amount of data produced by these devices will grow exponentially. Moreover, the high energy consumption for data transmission, the transmission latency, and the privacy and security constraints related to the managed sensitive data can make centralized cloud processing impracticable. Therefore, the need emerges to process data locally through devices capable of making inferences directly on the measurements collected by the sensors worn by patients. On the other hand, the resource constraints of such wearable devices and the computational complexity of exploited Machine Learning algorithms make this approach challenging. In this direction, Tiny Machine Learning and Federated Learning could represent the most appropriate solution for this challenge.