Medical embedded systems hold the promise to improve health outcomes, decrease isolation, reduce health disparities, and substantially reduce costs. In spite of their revolutionary potentials, these systems face a number of challenges in design and architecture that form stumbling blocks in their path to success. On one hand, as the sensor units continue to become more miniaturized, the underlying processing architectures demand for further miniaturization and power-efficiency to allow unobtrusive and long-term operation of the system. On the other hand, the data-intensive nature of continuous health monitoring requires efficient signal processing and data analytics techniques for real-time, scalable, reliable, accurate, and secure extraction of relevant information from an overwhelmingly large amount of data. In this paper, we present a data-processing-driven optimization and information extraction approach to address the problem of dynamic and power-aware feature selection for event classification applications using wearable sensors. Our results show that utilizing contextual information about users can reduce energy consumption of feature extraction module by 72.5% on average, compared to a static feature selection approach.