Endoscopic Sinus and Skull Base Surgeries (ESSBSs) is a challenging and potentially dangerous surgical procedure, and objective skill assessment is the key components to improve the effectiveness of surgical training, to re-validate surgeons' skills, and to decrease surgical trauma and the complication rate in operating rooms. Because of the complexity of surgical procedures, the variation of operation styles, and the fast development of new surgical skills, the surgical skill assessment remains a challenging problem. This work presents a novel Gaussian Process Learning-based heuristic automatic objective surgical skill assessment method for ESSBSs. Different with classical surgical skill assessment algorithms, the proposed method 1) utilizes the kinematic features in surgical instrument relative movements, instead of using specific surgical tasks or the statistics to assess skills in real-time; 2) provide informative feedback, instead of a summative scores; 3) has the ability to incrementally learn from new data, instead of depending on a fixed dataset. The proposed method projects the instrument movements into the endoscope coordinate to reduce the data dimensionality. It then extracts the kinematic features of the projected data and learns the relationship between surgical skill levels and the features with the Gaussian Process learning technique. The proposed method was verified in full endoscopic skull base and sinus surgeries on cadavers. These surgeries have different pathology, requires different treatment and has different complexities. The experimental results show that the proposed method reaches 100\% prediction precision for complete surgical procedures and 90\% precision for real-time prediction assessment.