Hybrid beamforming (HB) architectures are attractive for wireless communication systems with large antenna arrays because the analog beamforming stage can significantly reduce the number of RF transceivers and hence power consumption. In HB systems, channel estimation (CE) becomes challenging due to indirect access by the baseband processing to the communication channels and due to low SNR before beam alignment. Compressed sensing (CS) based algorithms have been adopted to address these challenges by leveraging the sparse nature of millimeter wave multi-input multi-output (mmWave MIMO) channels. In many CS algorithms for narrowband CE, the hybrid beamformers are randomly configured which does not always yield the low-coherence sensing matrices desirable for those CS algorithms whose recovery guarantees rely on coherence. In this paper, we propose a versatile deterministic HB codebook design framework for CS algorithms with coherence-based recovery guarantees to enhance CE accuracy. Simulation results show that the proposed design can obtain lower channel estimation error and higher spectral efficiency compared with random codebook for phase-shifter-, switch-, and lens-based HB architectures.