The Systems-on-Chip provide a large capacity for calculation and monitoring, so they are increasingly integrated into risky processes such as aeronautical and power generation systems. However, embedded systems are subject to degradation caused by wear, that can be accelerated by the often hostile environment. This paper proposes a method of failure prognosis of embedded systems based on the estimation of the temperature drift under reference operating conditions, then the modelling of the drift trend using a support vector regression model. The remaining useful life is estimated using the integral of the probability density function of the time to failure. Experimental results, evaluated by performance analysis techniques, show the effectiveness of the proposed approach.