Orthogonal time-frequency space (OTFS), which exhibits beneficial advantages in high-mobility scenarios, has been considered as a promising technology in future wireless communication systems. In this paper, a universal model for OTFS systems with generalized waveform has been developed. Furthermore, the average bit error probability (ABEP) upper bounds of the optimal maximum likelihood (ML) detector are first derived for OTFS systems with generalized waveforms. Specifically, for OTFS systems with the ideal waveform, we elicit the ABEP bound by recombining the transmitted signal and the received signal. For OTFS systems with practical waveforms, a universal ABEP upper bound expression is derived using moment generating-function (MGF), which is further extended to MIMO-OTFS systems. Numerical results validate that our theoretical ABEP upper bounds are concur with the simulation performance achieved by ML detectors.