Visible light communication (VLC) systems are expected to support a variety of applications, such as common-information broadcasting, real-time multimedia streaming, and large file downloading. Typically, these applications have different delay requirements. Hence, it is desirable to design high-performance coding scheme capable of supporting a wide range of delays but with acceptable hardware complexity. To achieve this, we propose a delay-tunable coding scheme for VLC systems based on block Markov superposition transmission of short non-binary low-density parity-check (NBLDPC) codes. The proposed coding scheme includes the following advantages: 1) it is easily configurable to fulfill different delay requirements while keeping the code rate constant; 2) it requires essentially the same hardware modules to implement the encoder/decoder as the involved short NBLDPC code; and 3) it can have a larger coding gain if a longer delay is tolerated. Numerical results are presented to show the expected tradeoff between delay and energy in a VLC system.