If we can get behind Harold Bloom's assertion that texts engender other texts–and thus that literary criticism, like fiction and poetry, is an artistic response to an earlier work–then Roland Barthes' S/Z is an excellent example of what Bloom had in mind. Rarely have I seen a critical work that so manhandles its source material. It is, after all, a 200-page essay on a 13-page story.
In the book, Barthes breaks down Balzac's story "Sarrasine" into 561 "lexias," which for him is a unit of any given text worth commenting on. Thus, for instance, lexia #2 reads: "I was deep in one of those daydreams," which then occasions a 2/3 of a page discussion. Clearly the lexias offer Barthes lots of room to let his imagination run loose, but he grants himself even more space: in S/Z one also finds a series of short digressions headed by Roman numerals and somewhat obscure titles (e.g. "XVIII. THE CASTRATO'S PROSPERITY"), these digressions bearing a certain tangential significance to the lexias they follow.
To say that in S/Z Barthes uses "Sarrasine" as license to write whatever he wants is overstating the case, but only slightly. The closest comparison I can think of at this point would be Pale Fire, where Kinbote's "criticism" (I use the term with hesitation) runs roughshod over Shade's poem, yet nonetheless remains captive to the logic of the source material.
One way you can tell Barthes' respect for the text he critiques is by how he continually imbues it with agency, as if it was an organism in its own right. This must be the first critical book I've read in which the text (here personified as "the discourse") actually becomes a subject, with certain needs and wants. E.g.:
. . . for the moment, probably due to the needs of the Antithesis, the discourse can only contrast the old man-machine with the new child-woman.
Essentially, in S/Z Barthes is picking apart "Sarrasine" into constituent pieces (although he's the first to acknowledge that the pieces he discovers are but one way of breaking up this piece of writing). A bit like an archaeologist, he's assigning each piece certain traits and organizing them along various schemas of his own creation. Thus, in S/Z he is attempting to create a response to "Sarrasine" that disrespects the narrative logic of the story (for example, he completely spoils the ending within the first pages, but in an analysis such as this "spoiler" really has no meaning), while simultaneously creating a different kind of critical logic. Thus, in my reading, it's a creative response to a source text, similar to how an author of fiction would reinvent a source text in his or her own writing (e.g. the use of Hawthorne and the Quixote in The New York Trilogy, or the use of the Odyssey in Ulysses).
The last thing I want to say about this book right now is that it's delightfully full of all kinds of the portentous statements that I love to encounter in Barthes. For instance: "Thus, realism (badly named, at any rate often badly interpreted) consists not in copying the real but in copying a depicted copy of the real." Or: "Beauty (unlike ugliness) cannot really be explained: in each part of the body it stands out, repeats itself, but it does not describe itself. Like a god (and as empty), it can only say: 'I am what I am.'" And lastly, what at this point must be my favorite, if only for its audacity:
And then: if literature and painting are no longer held in a hierarchical reflection, one being the rear-view mirror for the other, why maintain them any longer as objects at once united and separate, in short, classed together? Why not wipe out the difference between them (purely one of substance)? Why not forgo the plurality of the "arts" in order to affirm more powerful the plurality of "texts"?