Brain Machine Interfaces (BMI) have been developed as an alternative way to decode brain signals into control commands and communication devices. A typical BMI uses a computer to process EEG signals; however, current embedded PCs have enough computational resources for fully embedded BMI systems. In this work, the performance of the Odroid-xu4 embedded PC is evaluated as a processing and control device for BMI based on a 2-class motor imagery paradigm. Results show the best accuracy (82.1%) using SVM classifier and minimal processing times (0.11s) on the embedded device, which allows the development of a portable, low cost and trustworthy system.