Precise control of air-fuel ratio (AFR) is one of the most challenging tasks for lean-burn spark ignition engines control. The main problem arises due to the large time-varying delay in the engine operating envelope. In this paper, we present a sliding mode-based synthesis method to control AFR in order to increase fuel economy and decrease the tailpipe emissions. The time-varying delay dynamics is first estimated by the Padé approximation, which transfers the system into a system with parameter-varying non-minimum phase dynamics. Non-minimum phase characteristics restrict the application of conventional sliding mode control approach due to the unstable internal dynamics. The system dynamics is then rendered into the normal form to investigate the system unstable internal dynamics. A systematic approach is proposed to design a dynamic sliding manifold (DSM) in order to stabilize the unstable internal dynamics according to the desired output tracking error dynamics. Additionally, the proposed DSM provides the system with robustness against unmatched perturbation. The results of applying the proposed method on experimental data demonstrate the closed-loop system stability and a superior performance against time-varying delay, canister purge disturbances and measurement noise.