Parental Control functionalities currently included in Smart TVs, DBS decoders and Pay TV services are cumbersome to use, not flexible and, in the end, are seldom used. This paper presents a framework for dynamic enforcement of parental control policies on Smart TV contents. The framework exploits an extended version of the Usage Control model able to dynamically grant, suspend and resume the right to access a content, based on the content classification and on the age of people currently in front of the Smart TV. An accurate age estimation is performed by mean of a cascade of two deep learning networks. We present an implementation of the proposed framework in a smart home environment, showing both simulated and real experiments for accuracy in classification and performance analysis.