You do not have to put the R code in the input document; with knitr, you can separate your input document with the R script (think the reverse of Stangle).
There are several advantages of separating the main document with the R script(s), e.g., R code can be reusable across several documents, and you can run the R code continuously in a separate file (if you embed the code in the document, you often have to jump through texts); this feature is especially useful for LyX users, and it saves a huge amount of time since you do not have to re-compile the whole document to see the results; instead, you can tune your R code freely in another R session.
The function read_chunk() was designed for this feature. For example, if the R code is in 113-foo.R, you can use read_chunk('113-foo.R') to read the code into the input document in an early chunk. In the R script, you have to annotate the code with comments of the form ## ---- label, where label is the chunk label that you use in the report, e.g. if you have a line ## ---- test in the script, you should also have a chunk <<test>>= in the report, and knitr will match the labels and insert external code.
It is possible to use multiple external R scripts in an input document (just call read_chunk() for multiple times), or share a single script across multiple input documents (all of them read this script).
Note this function should only be used in an uncached chunk, e.g.
<<external-code, cache=FALSE>>=
read_chunk('foo-bar.R')
@
See the examples #113 in the knitr-examples repo.