This paper proposes a dynamic behavioral model for temperature variations of systems on chips (SoC) in embedded systems. We use identification techniques (ARMAX modeling) to construct a data-driven online temperature model that estimates the temperature according to the CPU and GPU frequencies, the used RAM and the power consumed by the chip. Furthermore, we used two the Recursive Least Squares (RLS) to estimate the parameters of the ARMAX model. This method allows us to update the parameters of the model online in case of a change in the system or its characteristics. Finally, we validate the temperature model and compare between booth estimation methods.