The deployment of TinyML models on edge resource-constrained devices introduces severe security challenges. Unlike cloud environments, physically accessible edge hardware exposes proprietary Neural Network models to extraction, reverse-engineering, and tampering. Traditional cryptographic defenses exceed the strict memory and processing budgets of MCU-class devices. Existing hardware-bound solutions rely on custom Inline Crypto Engines that lack flexibility, while commercial compiler-driven approaches assume massive on-chip memory to perform decryption at layer granularity. Consequently, a critical gap remains: no open-source framework integrates weight encryption with software-driven memory tiling at micro-granularity for severely constrained heterogeneous RISC-V platforms. This thesis addresses this gap by proposing an end-to-end secure deployment framework that proactively orchestrates cryptographic routines at the tile level.