In this work, we present a methodology to compress videos where human faces are the primary subject of interest. Our methodology comprises of detection of faces in each image frame within a given video, and blurring the remaining uninteresting part of the image, such as the image background. We apply an adaptive blurring kernel that dynamically adapts to the image characteristics, to gradually blur our the image in proportion to the distance of each given point of the image from its focus of interest (the face). We validate our work on two benchmark databases, namely the Talking Face Video and the Youtube Celebrities Face Dataset, as well as, a real-life user study towards user satisfaction. Our system can be applied in real-life facial video storage and network streaming, over and beyond the standard compression techniques that exist today.