Abstract : Modeling heat and mass transfer in two-phase flows with phase-change is crucial in many industrial studies including nuclear safety. Only averaged two-phase flow models can simulate such complex flows. Their accuracy depends in particular on closure laws for interfacial mass, momentum, and energy transfers that often rely on experimental correlations. Supporting averaged models, the goal of this thesis is to bring local closure information from finer simulations to consider 3D-effects and interfacial transfers more accurately. In this prospect, as direct resolution of the local balance equations is too expensive, we seek for a two-phase equivalent of Large Eddy Simulation (LES) in order to tackle simulations with enough bubbles to extract statistics needed in averaged models. Applying a spatial filter, we aim at modeling subgrid turbulence and interfacial transfers. The largest turbulent scales and interface deformations are captured because the filter size is chosen in-between the Kolmogorov scale and the bubble size. Because of the importance of the phase-change phenomena, this thesis extend the Interfaces and Subgrid Scales (ISS) model proposed by Toutant et al. (2009a) to non-material interfaces, i. e., with phase-change. Explicit interface tracking is valuable to accurately estimate interfacial transfers such as the phase-change rate. In the first part of this document, we establish a smeared-interface description of two-phase flows. Sub-grid transfers and interfacial deformations are modeled using the modified Bardina et al. (1983) scale-similarity hypothesis. Main modeling issues comes from (i) the specific interfacial velocity, (ii) the velocity and the temperature gradient discontinuities at the interface and (iii) the saturation condition of the interface. Models are validated using reference data from DNS. In the second up-scaling step, we transform interfacial subgrid models into source terms in the jump conditions in order to establish an equivalent discontinuous model thus benefitting from the knowledge acquired in DNS numerical methods. Transfers between phases are modified and the interfacial velocity is redefined considering the time evolution of curvature and the velocity jump at the interface. As a result, the normal momentum jump is modified. A tangential velocity jump is also introduced to cancel out the sub-resolution of the boundary layer. From a thermal point of view, the classical saturation condition is recovered ; the phase-change rate not only depends on the conduction heat flux but a contribution from the subgrid velocity and temperature correlations must be added to account for the poor resolution of the thermal boundary layer. As for single-phase LES, ISS modeling enables local-scale simulations of industrial configurations. It is the first step of a multi-scale approach towards turbulent bubbly flows. In this thesis, we illustrates how to bridge the gap between DNS and averaged descriptions from reference results obtained on condensing bubbles in a pseudo-turbulent subcooled liquid. Averaged quantities are compared with correlations for the condensation sink term used in the two-fluid model. We are able to underline the phase-change enhancement with increasing void fraction. It shows that this path could be used to improve the understanding of the strong two-way coupling between flow dynamics and interfacial heat transfers.