Reconfigurable intelligent surface (RIS) has been regarded as a promising tool to strengthen the quality of signal transmissions in non-orthogonal multiple access (NOMA) networks. This article introduces a heterogeneous network (HetNet) structure into RIS-aided NOMA multi-cell networks. A practical user equipment (UE) association scheme for maximizing the average received power is adopted. To evaluate system performance, we provide a stochastic geometry based analytical framework, where the locations of RISs, base stations (BSs), and UEs are modeled as homogeneous Poisson point processes (PPPs). Based on this framework, we first derive the closed-form probability density function (PDF) to characterize the distribution of the reflective links created by RISs. Then, both the exact expressions and upper/lower bounds of UE association probability are calculated. Lastly, the analytical expressions of the signal-to-interference-plus-noise-ratio (SINR) and rate coverage probability are deduced. Additionally, to investigate the impact of RISs on system coverage, the asymptotic expressions of two coverage probabilities are derived. The theoretical results show that RIS length is not the decisive factor for coverage improvement. Numerical results demonstrate that the proposed RIS HetNet structure brings significant enhancement in rate coverage. Moreover, there exists an optimal combination of RISs and BSs deployment densities to maximize coverage probability.