The ubiquitous operation of mobile and embedded devices has given an impetus to the development of sensing systems. Most applications on edge devices rely heavily on sensor inputs. Surveillance devices and autonomous vehicles often require high-frequency video sensor data for security provisions and decision making. Malicious applications on an edge device can re-architect video frames via attacks such as noise-addition and blurring. This augments the need for a chain of trust to be established from the data capture to its delivery such that target applications can establish sensor data authenticity and fidelity. The key contribution of this paper is securing high frequency video streams using memory and compute constrained hardware security extensions in real-time by tuning memory and compute intensive computer vision algorithms through domain specific optimizations. We put forth the consequent challenges and constraints of applying complex image processing at the edge devices while utilizing a limited secure hardware storage. The preliminary evaluation is performed on an off-the-shelf embedded device to outline the credibility of our proposed framework.