Channel estimation and data detection constitute a pair of pivotal modules in multiple-input multiple-output (MIMO) communication systems, where achieving accurate channel estimation is particularly important for large-scale MIMO communications. However, more accurate channel estimation requires more resources. Hence, we investigate the joint optimization of the channel estimator, transmit precoder and receiver in large-scale MIMO systems. In contrast to the classic signal processing philosophy, the joint optimization aims for solving two equations in the face of realistic channel estimation and data transmission imperfections. Closed-form solutions are derived for a pair of schemes. For the first one, the joint optimization consists of the three procedures of channel estimation, data estimation and channel refinement. In this method, the estimated data symbols are also harnessed as pilots, based on which the channel estimation performance is improved. As for the second scheme, since data estimation is our final goal and channel estimation is only an intermediate step, the channel estimation procedure is substituted into the data estimation regime without deriving an explicit solution for the estimated channel. Given our objective of optimizing the data estimation performance, the channel estimator and data transceiver are jointly optimized, and the intricate linkages between these two methods are discussed. Finally, several numerical results are provided for demonstrating the performance advantages over the traditional designs.