This study proposes a Gaussian Process Regression (GPR) model with a custom composite kernel for age prediction from normal electroencephalogram (EEG) signals. Using the TUH Abnormal EEG Corpus from the Temple University EEG dataset, the model combines Radial Basis Function (RBF) and Matérn kernels to capture both global and local variations in EEG features. The prediction framework demonstrates a mean absolute error of 5.252 years and shows higher accuracy for middle-aged individuals. The composite kernel enables flexible adaptation to EEG signal characteristics across age ranges. The approach highlights differences in prediction performance across age groups and suggests clinical applicability for EEG-based age estimation. The findings demonstrate the potential of combining feature engineering and kernel-based learning for age prediction from normal EEG signals.