The faults diagnosis of machines consists of feature extraction and classification of faults. The fault diagnosis process is based on the fact that each fault in a machine has a unique vibrational feature. Every industry wants a compact device or embedded system of low cost for the early faults diagnosis of machinery. This paper presents the platform that is focused at the embedded system design for the early faults diagnosis of machine(s) and classification of faults. We performed our experiment on the test rig apparatus and collected the vibration signals of four states of machine, those were: normal state, cracking state, offset pulley state and wear state. In segmentation, we use the empirical mode decomposition (EMD) technique. For classification purpose, we are using k-nearest neighbors (K-NN). Achievement of this research is that embedded system design for the classification of different faults in the machines. The overall accuracy of our experiment is 91.5%.