Efficient clutter filtering for pulsed radar systems remains an open issue when employing pulse-to-pulse modulation and irregular pulse interval waveforms within the coherent processing interval. The range and Doppler domain should be jointly processed for effective filtering leading to a large computational overhead. In this paper, the joint domain filtering is performed by constructing a clutter projection matrix, also known as the projected non-identical multiple pulse compression (NIMPC) method. The paper extends the projected NIMPC filter to irregular pulse interval waveforms. Additionally, a kernel-based regularization will be introduced to tackle the ill-conditioning of the matrix inverse of the NIMPC method. The regularization is based on a model of the second-order statistics of the clutter. Moreover, a computationally efficient algorithm is formulated based on fast Fourier transforms and the projected conjugate gradient method. Through a Monte Carlo study it is demonstrated that the proposed kernelized filtering outperforms the projected NIMPC in clutter filtering.