The synergy of full-duplex (FD) radio and mil-limeter-wave (mmWave) communications can be jointly beneficial, as FD can enhance the spectral efficiency of a point-to-point link, while the prominent properties of mmWave communications can mitigate the overall interference. However, the FD operation in the context of large-scale networks leads to an increased multi-user interference, imposing a trade-off between improving the average spectral efficiency and increasing the aggregate interference. Furthermore, although the increase of multi-user interference due to the FD operation is harmful for information decoding, it can be beneficial for energy harvesting, and hence, the trade-off between successful information decoding and average harvested energy emerges. In this article, we describe the application of stochastic geometry to model and design FD-enabled mmWave communications in the context of large-scale heterogeneous networks. Based on the proposed framework, we assess the performance of heterogeneous FD-mmWave cellular networks under two user location-based classifications, namely cell-center users (CCUs) and cell-edge users (CEUs), and evaluate the trade-offs imposed by the FD operation. Numerical results demonstrate that the half-duplex operation of the CEUs is preferable for achieving better network performance, in contrast to the CCUs for which FD operation is more efficient.