This paper presents an anthropocentric approach in human-machine interaction in the area of self-regulated e-learning. In an attempt to enhance communication mediated through computers for pedagogical use we propose the incorporation of an intelligent emotional agent that is represented by a synthetic character with multimedia capabilities, modelled to imitate human behaviour. The agent is aiming to provide cognitive support to users with learning difficulties and attention disorders and is designed to accommodate self regulated learning elements. We review the basic principles of self regulated learning which, in turn, act as a basis for designing and implementing our system. Kolb's learning cycle is used to provide a framework upon which agents' pedagogical behaviour is constructed. A study between 24 students from higher education with learning difficulties and attention disorders is presented. The learning particularities of this special group that contradict with the principles of self regulated learning are reported. The study refers to students in higher education, in the domain of information technology. The analysis of results indicates that emotional agents improve communication between users of the particular learning group and learning environments by providing cognitive support through behavioural communication, compared to agents with neutral behaviour.