The objective of the proposed research is to address the increasing deployment of situation-awareness applications, such as autonomous vehicles and sensor networks, which necessitate efficient and low-latency data processing solutions. Traditional cloud computing infrastructures struggle to meet these demands due to significant latency and bandwidth constraints. This thesis proposes a novel system architecture designed specifically for situation-awareness applications at the edge. Leveraging serverless computing models, particularly Function as a Service, this architecture optimizes machine learning model deployment, elevates accelerators as first-class citizens in scheduling, and ensures efficient data management. Key contributions include PerformanceDNA, an automated tool for optimal ML model deployment; MicroEdge, a multi-tenant edge cluster system for efficient accelerator sharing; and STateFlow, a data management layer that enhances data locality and correctness. Together, these innovations aim to reduce developer burden, improve resource utilization, meet stringent service level objectives, and maintain the integrity of situation-awareness applications, thereby advancing the state-of-the-art in edge computing.