Many-core computing architectures such as GPUs are becoming first-class computing engines for accelerating important workloads including machine learning, genome analysis, graph processing, databases, and scientific simulation. Unfortunately, existing programming models deployed on many-core architecture are designed and optimized largely based on the assumptions for single-core/multi-core architectures. This talk focuses on the unique challenges that arise when programming many-core and how these challenges affect the design principles of a programming system. Specifically, I will talk about three fundamental components that are critical to the performance and the secure execution of programs running on a many-core processor: 1) enabling communication-aware program development and transformation, (2) integrating concurrency optimization into program tuning framework, and (3) supporting dynamic information flow tracking for data protection during program execution.