Side channel attacks extracting secrets carried by the power consumption variations or electromagnetic emissions in embedded devices are a consolidated threat to the security of edge computing systems. Such attacks either employ a synthetic model for the device behaviour to predict secret-dependent components of the measured power consumption (non-profiled attacks), or obtain such a model in a data-driven fashion (profiled attacks). Protections against both profiled and non-profiled attacks are characterized by a significant overhead, typically one or two orders of magnitude in computation time, and a comparatively significant engineering effort to deploy them. Furthermore, such protections are designed to hinder one of the two aforementioned attack strategies. In this work, we propose a compiler-based methodology to automate the application of a comparatively inexpensive countermeasure able to hinder both profiled and unprofiled attacks. We experimentally validate our approach employing the AES symmetric cipher as our case study, and a Cortex-M4 based microcontroller as the target device. Our solution increases the Measurements-to-Disclose security metric by at least \(5000\times \) in an attacker-optimal scenario, and proves to be immune to Bayesian template- and SVM-based profiled attacks.