You can install NIMBLE as an R package from CRAN in the usual way.

You can install from the source file using
   R CMD INSTALL nimble_1.3.0.tar.gz
or
  install.packages("nimble_1.3.0.tar.gz", repos = NULL)

You can also install the package directly from the NIMBLE repository 
  install.packages("nimble", repos = "https://r-nimble.org")

There are two configuration options:  --enable-dylib and --with-eigen.
--enable-dylib is supported on UNIX/Linux/OS X platforms,
i.e. not Windows.

 --enable-dylib=false/true  controls whether we compile the shared C++ code that is used
   across different NIMBLE models as a dynamic library, libnimble.so, and then use this one copy of the code
   by linking against it. Otherwise, we create a static library, libnimble.a, and include this in the DLL for each compiled model and algorithm, essentially copying the code. By default, --enable-dylib=true for Linux but --enable-dylib=false for OS X. If building from source for OS X, we recommend setting --enable-dylib=true to reduce the size of the shared libraries built when compiling models and algorithms. 

 --with-eigen=/path/to/dir   - this specifies the full path to a directory containing the 
    header files for the Eigen library. This directory must contain the Eigen/ sub-directory 
    By default, we will use the Eigen distribution included with the package.

We can specify any or all of these with
  R CMD INSTALL nimble_1.3.0.tar.gz  --configure-args="--enable-dylib=true --with-eigen=/Users/duncan/local"
or
  install.packages("nimble", configure.args = "--enable-dylib=true --with-eigen=/Users/duncan/local")

Note that if you install on MacOS from the source file provided by CRAN, you will need to install the GNU Fortran compiler package, as discussed at https://mac.r-project.org/tools. In contrast, if you install from source from r-nimble.org, we leave out $FLIBS from src/Makevars.in so you won't need any gfortran package installed.