Abstract: Detecting the Cosmic Microwave Background B-mode polarisation is one of the major challenges of nowadays cosmology. This measurement indeed appears as the most powerful way to place constraints on inflation, this process of extreme accelerated expansion that would have occurred during the first few moments of the Universe. The QUBIC project, one of the many experiments dedicated to this purpose, is based on a novel instrumental technique we call bolometric interferometry. The work described in this thesis attempts to establish the instrument design. We introduce a mathematical formalism that shows how the observables of interest can be measured. We then obtain an analytical formula giving the instrument sensitivity and derive the requirements that its components must satisfy. We show in particular that the receivers (horns) distribution must be highly redundant in order for the instrument to achieve an optimal sensitivity. We study the bandwidth smearing effect due to bolometers' wide spectral band and estimate the resulting sensitivity loss. We also introduce alternative designs of bolometric interferometers ; one of them (with rotating half-wave plate and no controlled phase-shifters) has become the standard design of the QUBIC instrument.We finally present a promising self-calibration procedure that, we think, may be regarded in the future as one of the main arguments in favor of bolometric interferometry.