Accurate estimation of solar radiation components of a specific location has been one of the most important issues of solar energy applications. In this paper, a new approach, named Weighted Gaussian Process Regression (WGPR), is developed for multi-step ahead forecasting of daily global and direct horizontal solar radiation components in Saharan climate. The WGPR is tested using global and direct solar radiation data recorded over three years (2013–2015) in a semi-arid region in Algeria. It consists of forecasting 10-steps ahead for both components with automatic selection of relevant climatic data. In this respect two different architectures of WGPR are proposed, WGPR Parallel Forecasting Architecture (WGPR-PFA) and WGPR Cascade Forecasting Architecture (WGPR-CFA). The proposed approach proved to be effective with respect to the basic GPR in terms of accuracy and processing time for daily global and direct solar radiation forecasting. Forecasting with WGPR-CFA led to error RMSE = 3.18 (MJ/m²) and correlation coefficient r² = 85.85 (%) for the 10th daily global horizontal radiation, and RMSE = 5.23 (MJ/m²) and correlation coefficient r² = 56.21(%) for 10th daily direct horizontal radiation. The achieved results specify that the developed WGPR approach can be adjudged as an efficient machine learning model for accurate forecasting of solar radiation components.