Dynamics of self-assembly and structural transitions in amphiphilic systems play a key role in various technological and biological processes. Understanding these processes is of immense importance, for example, in development of new nanostructured materials and drug delivery methods, as well as in deciphering intracellular transport mechanisms. One of the challenges toward fundamental understanding of self-assembly processes is that theoretical and computational investigations are complicated by a non-trivial interplay between different length- and time-scales. In this talk we present a development of a multi-scale approach to study the process of self-assembly. This approach links molecular dynamics simulations with a stochastic model that explicitly takes into account a small number of the relevant degrees of freedom (reaction coordinates) and treats all other degrees of freedom as a thermal noise. This stochastic model facilitates the analysis of complex self-assembly processes that are outside the reach of direct molecular dynamics simulations. We have applied this model to investigate dynamics of micelle formation and disintegration. The process of formation (disintegration) of a micelle is assumed to occur by a series of elementary steps, including addition (removal) of a single monomer to (from) the micelle, fusion of small amphiphilic clusters (premicelles) into the micelle, and fission of the micelle into small clusters. Each of these elementary steps is modeled by a Langevin equation. The parameters of the Langevin equation are obtained from a series of short-scale molecular dynamics simulations with constrained reaction coordinates. The most relevant choice of a reaction coordinate for the above processes is the distance between the micelle and the monomer being added or removed and the distance between the premicelles undergoing fusion or fission. The Langevin equation for a single reaction coordinate adequately describes the processes of a monomer removal from a micelle and micellar fission. However, we observe that the addition of a surfactant monomer to a micelle cannot be adequately described by a single degree of freedom. In this process, additional reaction coordinates, such as the microstructure of the micelle and the orientation of the monomer play an important role. Similarly, during fusion of premicelles, their microstructures manifest additional reaction coordinates since they significantly affect the fusion dynamics. The parameters of the Langevin equation for such processes are difficult to obtain from constraint simulations as it is not feasible to constrain the microstructure of a micelle. Hence we use short-scale unconstrained molecular dynamics simulations to develop a multi-dimensional Langevin equation for dynamics of monomer addition to a micelle and fusion of premicelles. In conclusion, we discuss possible extensions of the developed approach to model more complex processes, such as kinetics of formation of nanopartices in reverse micelles, rheology of solutions of worm-like micelles, and fusion of lipid bilayers.